diff --git a/.agents/skills/aiq-configure-workflow/assets/config-scaffold.yml b/.agents/skills/aiq-configure-workflow/assets/config-scaffold.yml
index 8539744f2..a987d763c 100644
--- a/.agents/skills/aiq-configure-workflow/assets/config-scaffold.yml
+++ b/.agents/skills/aiq-configure-workflow/assets/config-scaffold.yml
@@ -17,25 +17,29 @@ general:
   # Web/API: copy general.front_end from config_web_default_llamaindex.yml
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -68,11 +72,11 @@ functions:
 
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
 
   deep_research_agent:
     _type: deep_research_agent
diff --git a/.agents/skills/aiq-configure-workflow/references/config-profiles.md b/.agents/skills/aiq-configure-workflow/references/config-profiles.md
index d80f3f0ad..ae5c82b5b 100644
--- a/.agents/skills/aiq-configure-workflow/references/config-profiles.md
+++ b/.agents/skills/aiq-configure-workflow/references/config-profiles.md
@@ -17,7 +17,7 @@ the file from `configs/`. For the full nine-profile table, see
 3. **Model family**
    - Nemotron → most profiles
    - GPT-5.2 orchestration/planning/writing → `config_frontier_models.yml` (`OPENAI_API_KEY`)
-   - Default split → existing Super profiles for intent/shallow, with Ultra for deep research and a larger writer budget
+   - Default split → Nemotron 3.5 Lightning for intent/shallow; Ultra for clarification/deep research, with a larger writer budget
 
 4. **Optional features** — copy blocks from:
    - Guardrails → `config_web_default_guardrails.yml`
diff --git a/README.md b/README.md
index 8b2f939a1..31de51b62 100644
--- a/README.md
+++ b/README.md
@@ -108,8 +108,8 @@ The checked-in default CLI and web profiles use these core components:
 
 - [NVIDIA NeMo Agent Toolkit 1.8.0](https://docs.nvidia.com/nemo/agent-toolkit/latest/)
 - [LangChain Deep Agents](https://docs.langchain.com/oss/python/deepagents/overview) 0.6.5 or newer
-- [NVIDIA Nemotron 3 Super](https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b) for intent classification and shallow research in the default profiles
-- [NVIDIA Nemotron 3 Ultra](https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b) for clarification and every deep-research role in the default profiles
+- NVIDIA Nemotron 3.5 Lightning for intent classification and shallow research in the default profiles
+- [NVIDIA Nemotron 3 Ultra](https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b) for clarification and every deep-research role
 - [Google Gemma 4 31B IT](https://build.nvidia.com/google/gemma-4-31b-it) (document summary, if used)
 - [NVIDIA Nemotron 3 Embed 1B](https://build.nvidia.com/nvidia/nemotron-3-embed-1b) (embedding model for the knowledge layer, if used)
 - [NVIDIA Nemotron 3 Nano Omni 30B A3B Reasoning](https://build.nvidia.com/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning) (vision-language model for the LlamaIndex knowledge layer, if used)
@@ -152,7 +152,7 @@ When using [NVIDIA API Catalog](https://build.nvidia.com/) (the default), infere
 
 | Component | Default Model | Self-Hosted Hardware Reference |
 |-----------|---------------|-------------------------------|
-| LLM (intent classifier, shallow researcher) | `nvidia/nemotron-3-super-120b-a12b` | [Nemotron 3 Super](https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b) |
+| LLM (intent classifier, shallow researcher) | `nvidia/nemotron-3.5-lightning-30b-a3b` | [Nemotron 3.5 Lightning](https://build.nvidia.com/nvidia/nemotron-3.5-lightning-30b-a3b/modelcard) |
 | LLM (clarifier and all deep-research roles) | `nvidia/nemotron-3-ultra-550b-a55b` | [Nemotron 3 Ultra](https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b) |
 | Document summary (optional) | `google/gemma-4-31b-it` | [Gemma 4 31B IT](https://build.nvidia.com/google/gemma-4-31b-it) |
 | Text embedding | `nvidia/nemotron-3-embed-1b` | [NeMo Retriever embedding support matrix](https://docs.nvidia.com/nim/nemo-retriever/text-embedding/latest/support-matrix.html) |
@@ -289,17 +289,17 @@ The `configs/` directory holds YAML workflow configs that define agents, tools,
 
 | Config | Models | Description |
 |--------|--------|-------------|
-| `config_cli_default.yml` | Nemotron 3 Super/Ultra | CLI chat pipeline with Tavily and clarification; no knowledge backend. Paper search is a commented opt-in. |
-| `config_web_default_llamaindex.yml` | Nemotron 3 Super/Ultra; Gemma 4 summary | Default web/API chat pipeline with LlamaIndex/ChromaDB and Tavily. Paper search is commented out. |
-| `config_web_frag.yml` | Nemotron 3 Super/Ultra | Web/API and Helm base with Foundational RAG plus Tavily. Requires separately deployed RAG query and ingestion services. |
-| `config_web_opensearch.yml` | Nemotron 3 Super/Ultra; Nemotron 3 Embed | Web/API with built-in OpenSearch knowledge retrieval plus Tavily; supports self-hosted, `es`, and `aoss` authentication modes. |
-| `config_web_azure_ai_search.yml` | Nemotron 3 Super/Ultra; Nemotron 3 Embed | Web/API with Azure AI Search knowledge retrieval plus Tavily; supports API-key and Azure identity authentication. |
-| `config_frontier_models.yml` | GPT Sol/Luna; Nemotron 3 Super; Gemma 4 summary | LlamaIndex profile using GPT Sol/Luna for deep research and Super for intent/shallow. Requires `OPENAI_API_KEY` and `NVIDIA_API_KEY`. |
-| `config_web_default_guardrails.yml` | Nemotron 3 Super/Ultra; Gemma 4 summary | LlamaIndex profile with workflow Guardrails explicitly attached, shallow-agent Guardrails dynamically attached through `workflow_functions`, and async deep-agent Guardrails applied by the AI-Q runner from the same target configuration. |
-| `config_web_frag_mcp_auth.yml` | Nemotron 3 Super/Ultra | Foundational RAG plus an opt-in protected per-user OAuth MCP source example. Requires a real MCP endpoint and shared token store. |
+| `config_cli_default.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra | CLI chat pipeline with Tavily and clarification; no knowledge backend. Paper search is a commented opt-in. |
+| `config_web_default_llamaindex.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra; Gemma 4 summary | Default web/API chat pipeline with LlamaIndex/ChromaDB and Tavily. Paper search is commented out. |
+| `config_web_frag.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra | Web/API and Helm base with Foundational RAG plus Tavily. Requires separately deployed RAG query and ingestion services. |
+| `config_web_opensearch.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra; Nemotron 3 Embed | Web/API with built-in OpenSearch knowledge retrieval plus Tavily; supports self-hosted, `es`, and `aoss` authentication modes. |
+| `config_web_azure_ai_search.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra; Nemotron 3 Embed | Web/API with Azure AI Search knowledge retrieval plus Tavily; supports API-key and Azure identity authentication. |
+| `config_frontier_models.yml` | GPT Sol/Luna; Nemotron 3.5 Lightning; Gemma 4 summary | LlamaIndex profile using GPT Sol/Luna for deep research and Nemotron 3.5 Lightning for intent/shallow. Requires `OPENAI_API_KEY` and `NVIDIA_API_KEY`. |
+| `config_web_default_guardrails.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra; Gemma 4 summary | LlamaIndex profile with workflow Guardrails explicitly attached, shallow-agent Guardrails dynamically attached through `workflow_functions`, and async deep-agent Guardrails applied by the AI-Q runner from the same target configuration. |
+| `config_web_frag_mcp_auth.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra | Foundational RAG plus an opt-in protected per-user OAuth MCP source example. Requires a real MCP endpoint and shared token store. |
 | `config_domain_routing_and_skills.yml` | Nemotron 3 Ultra; Gemma 4 summary | Direct deep-research profile with domain routing, DuckDuckGo news, Polymarket, enabled Serper paper search, LlamaIndex, built-in skills, and a fresh per-job Modal sandbox. |
-| `config_openshell.yml` | Nemotron 3 Super/Ultra; Gemma 4 summary | Experimental web/API skills profile with artifact capture, fail-closed policy attestation, and one OpenShell sandbox per deep-research job. |
-| `config_mcp.yml` | Nemotron 3 Super/Ultra | Standalone MCP server. Public NIM + Tavily research with PostgreSQL-backed stateless submit/poll/report. Requires `NVIDIA_API_KEY`, `TAVILY_API_KEY`, and `AIQ_CHECKPOINT_DB`. |
+| `config_openshell.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra; Gemma 4 summary | Experimental web/API skills profile with artifact capture, fail-closed policy attestation, and one OpenShell sandbox per deep-research job. |
+| `config_mcp.yml` | Nemotron 3.5 Lightning; Nemotron 3 Ultra | Standalone MCP server. Public NIM + Tavily research with PostgreSQL-backed stateless submit/poll/report. Requires `NVIDIA_API_KEY`, `TAVILY_API_KEY`, and `AIQ_CHECKPOINT_DB`. |
 
 ## Ways to Run the Agents
 
diff --git a/configs/config_cli_default.yml b/configs/config_cli_default.yml
index 76913b37a..a89889af8 100644
--- a/configs/config_cli_default.yml
+++ b/configs/config_cli_default.yml
@@ -16,25 +16,29 @@ general:
     #     project: nvidia-aiq
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -115,7 +119,7 @@ functions:
   # =========================================================================
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     # tools: omitted -> inherits all from data_source_registry
     # exclude_tools: []
     # llm_timeout: 90  # optional; seconds for intent LLM call (default 90)
@@ -131,7 +135,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     # tools: omitted -> inherits all from data_source_registry
     exclude_tools:
       - advanced_web_search_tool
diff --git a/configs/config_frontier_models.yml b/configs/config_frontier_models.yml
index f66f21123..ebbf623eb 100644
--- a/configs/config_frontier_models.yml
+++ b/configs/config_frontier_models.yml
@@ -51,25 +51,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.1
-    top_p: 0.3
-    max_tokens: 16384
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
+    top_p: 0.7
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -137,7 +141,7 @@ functions:
 
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     verbose: true
     tools:
       - web_search_tool
@@ -156,7 +160,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     verbose: true
     tools:
       - web_search_tool
diff --git a/configs/config_mcp.yml b/configs/config_mcp.yml
index 69b0019bc..d4bff8770 100644
--- a/configs/config_mcp.yml
+++ b/configs/config_mcp.yml
@@ -26,25 +26,29 @@ general:
         level: INFO
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -95,7 +99,7 @@ functions:
 
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
 
   # Defined for workflow compatibility but disabled at the workflow boundary.
   clarifier_agent:
@@ -107,7 +111,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     exclude_tools:
       - advanced_web_search_tool
     max_llm_turns: 10
diff --git a/configs/config_openshell.yml b/configs/config_openshell.yml
index c48bc9977..e0f3384df 100644
--- a/configs/config_openshell.yml
+++ b/configs/config_openshell.yml
@@ -28,25 +28,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -120,7 +124,7 @@ functions:
 
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     verbose: true
 
   clarifier_agent:
@@ -132,7 +136,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     exclude_tools:
       - advanced_web_search_tool
     verbose: true
diff --git a/configs/config_web_azure_ai_search.yml b/configs/config_web_azure_ai_search.yml
index acc5331eb..524693b6a 100644
--- a/configs/config_web_azure_ai_search.yml
+++ b/configs/config_web_azure_ai_search.yml
@@ -47,25 +47,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -187,7 +191,7 @@ functions:
   # =========================================================================
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     # tools: omitted -> inherits all from data_source_registry
     # exclude_tools: []
     verbose: true
@@ -205,7 +209,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     # tools: omitted -> inherits all from data_source_registry
     exclude_tools:                     # Remove advanced variant; shallow uses web_search_tool
       - advanced_web_search_tool
diff --git a/configs/config_web_default_guardrails.yml b/configs/config_web_default_guardrails.yml
index a8d43f3c9..7ebe273b7 100644
--- a/configs/config_web_default_guardrails.yml
+++ b/configs/config_web_default_guardrails.yml
@@ -28,25 +28,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -225,7 +229,7 @@ functions:
 
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     verbose: true
 
   clarifier_agent:
@@ -237,7 +241,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     exclude_tools:
       - advanced_web_search_tool
     verbose: true
diff --git a/configs/config_web_default_llamaindex.yml b/configs/config_web_default_llamaindex.yml
index 872497407..f860a241d 100644
--- a/configs/config_web_default_llamaindex.yml
+++ b/configs/config_web_default_llamaindex.yml
@@ -47,25 +47,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -188,7 +192,7 @@ functions:
   # =========================================================================
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     # tools: omitted -> inherits all from data_source_registry
     # exclude_tools: []
     verbose: true
@@ -204,7 +208,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     # tools: omitted -> inherits all from data_source_registry
     exclude_tools:                     # Remove advanced variant; shallow uses web_search_tool
       - advanced_web_search_tool
diff --git a/configs/config_web_frag.yml b/configs/config_web_frag.yml
index 99f593ce5..8f3c97599 100644
--- a/configs/config_web_frag.yml
+++ b/configs/config_web_frag.yml
@@ -49,25 +49,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -158,7 +162,7 @@ functions:
   # =========================================================================
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     # tools: omitted -> inherits all from data_source_registry
     # exclude_tools: []
 
@@ -173,7 +177,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     # tools: omitted -> inherits all from data_source_registry
     exclude_tools:
       - advanced_web_search_tool
diff --git a/configs/config_web_frag_mcp_auth.yml b/configs/config_web_frag_mcp_auth.yml
index 780b91882..9d63fe98d 100644
--- a/configs/config_web_frag_mcp_auth.yml
+++ b/configs/config_web_frag_mcp_auth.yml
@@ -56,25 +56,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -207,7 +211,7 @@ functions:
   # =========================================================================
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     # tools: omitted -> inherits all from data_source_registry
     # exclude_tools: []
 
@@ -222,7 +226,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     # tools: omitted -> inherits all from data_source_registry
     exclude_tools:
       - advanced_web_search_tool
diff --git a/configs/config_web_opensearch.yml b/configs/config_web_opensearch.yml
index 7b7452fe7..4eeedbeeb 100644
--- a/configs/config_web_opensearch.yml
+++ b/configs/config_web_opensearch.yml
@@ -48,25 +48,29 @@ general:
         - "*"
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -165,7 +169,7 @@ functions:
   # =========================================================================
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     # tools: omitted -> inherits all from data_source_registry
     # exclude_tools: []
 
@@ -180,7 +184,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     # tools: omitted -> inherits all from data_source_registry
     exclude_tools:
       - advanced_web_search_tool
diff --git a/docs/notebooks/0_Getting_Started_with_AIQ.ipynb b/docs/notebooks/0_Getting_Started_with_AIQ.ipynb
index 56c04385d..8fc5c72ce 100644
--- a/docs/notebooks/0_Getting_Started_with_AIQ.ipynb
+++ b/docs/notebooks/0_Getting_Started_with_AIQ.ipynb
@@ -71,7 +71,7 @@
     "The following are used by this project:\n",
     "\n",
     "- [NVIDIA NeMo Agent toolkit](https://docs.nvidia.com/nemo/agent-toolkit/latest/)\n",
-    "- [NIM of nvidia/nemotron-3-super-120b-a12b](https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b) for intent classification and shallow research\n",
+    "- [NIM of nvidia/nemotron-3.5-lightning-30b-a3b](https://build.nvidia.com/nvidia/nemotron-3.5-lightning-30b-a3b/modelcard) for intent classification and shallow research\n",
     "- [NIM of nvidia/nemotron-3-ultra-550b-a55b](https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b) for every deep-research role\n",
     "- [NIM of nvidia/nemotron-3-embed-1b](https://build.nvidia.com/nvidia/nemotron-3-embed-1b) (Optional)\n",
     "- [NIM of nvidia/nemotron-3-nano-omni-30b-a3b-reasoning](https://build.nvidia.com/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning) (Optional)\n",
@@ -94,7 +94,7 @@
     "If you are deploying this blueprint without the NVIDIA RAG knowledge layer backend, there are **no GPU requirements** for this blueprint on its own.\n",
     "\n",
     "- **Self-hosting NIMs:**\n",
-    "To self-host the default LLM stack, see the model cards for `nvidia/nemotron-3-super-120b-a12b` and `nvidia/nemotron-3-ultra-550b-a55b`.  \n",
+    "See the model cards for current deployment options for `nvidia/nemotron-3.5-lightning-30b-a3b` and the deep-research model `nvidia/nemotron-3-ultra-550b-a55b`.  \n",
     "If using NVIDIA RAG knowledge layer backend, please see [NVIDIA RAG Documentation Pages](https://docs.nvidia.com/rag/latest/support-matrix.html)\n",
     "\n"
    ]
@@ -303,25 +303,29 @@
     "        level: INFO\n",
     "\n",
     "llms:\n",
-    "  nemotron_llm_intent:\n",
+    "  nemotron_lightning_intent_llm:\n",
     "    _type: nim\n",
-    "    model_name: nvidia/nemotron-3-super-120b-a12b\n",
+    "    model_name: nvidia/nemotron-3.5-lightning-30b-a3b\n",
     "    base_url: \"https://integrate.api.nvidia.com/v1\"\n",
-    "    temperature: 0.5\n",
+    "    api_key: ${NVIDIA_API_KEY}\n",
+    "    temperature: 0.1\n",
     "    top_p: 0.9\n",
-    "    max_tokens: 4096\n",
+    "    max_tokens: 1024\n",
     "    num_retries: 5\n",
+    "    parallel_tool_calls: false\n",
     "    chat_template_kwargs:\n",
-    "      enable_thinking: true\n",
+    "      enable_thinking: false\n",
     "\n",
-    "  nemotron_super_llm:\n",
+    "  nemotron_lightning_agent_llm:\n",
     "    _type: nim\n",
-    "    model_name: nvidia/nemotron-3-super-120b-a12b\n",
+    "    model_name: nvidia/nemotron-3.5-lightning-30b-a3b\n",
     "    base_url: \"https://integrate.api.nvidia.com/v1\"\n",
-    "    temperature: 0.7\n",
+    "    api_key: ${NVIDIA_API_KEY}\n",
+    "    temperature: 0.2\n",
     "    top_p: 0.7\n",
-    "    max_tokens: 65536\n",
+    "    max_tokens: 8192\n",
     "    num_retries: 5\n",
+    "    parallel_tool_calls: false\n",
     "    chat_template_kwargs:\n",
     "      enable_thinking: true\n",
     "\n",
@@ -362,13 +366,13 @@
     "\n",
     "  intent_classifier:\n",
     "    _type: intent_classifier\n",
-    "    llm: nemotron_llm_intent\n",
+    "    llm: nemotron_lightning_intent_llm\n",
     "    tools:\n",
     "      - web_search_tool\n",
     "\n",
     "  shallow_research_agent:\n",
     "    _type: shallow_research_agent\n",
-    "    llm: nemotron_super_llm\n",
+    "    llm: nemotron_lightning_agent_llm\n",
     "    tools:\n",
     "      - web_search_tool\n",
     "\n",
@@ -401,7 +405,7 @@
     "\n",
     "**`general.front_end`** wires the Web API (AI-Q API Worker), async job store, and CORS for the chat UI.\n",
     "\n",
-    "**`llms`** retains Nemotron Super for intent and shallow research, and uses role-specific Nemotron Ultra profiles for every deep-research role.\n",
+    "**`llms`** wires Nemotron 3.5 Lightning for intent and shallow research, and uses role-specific Nemotron Ultra profiles for every deep-research role.\n",
     "\n",
     "**`functions`** register tools (Tavily web search, plus advanced search), the intent classifier, and the shallow and deep research agents with their assigned tools and LLMs.\n",
     " \n",
@@ -800,13 +804,13 @@
     "\n",
     "You can swap the LLM provider in `config_simple_researcher.yml` with any OpenAI-compatible API.\n",
     "\n",
-    "For example, to use [Together.ai](https://www.together.ai/) with the [Nemotron model](https://www.together.ai/models/nvidia-nemotron-3-super) — sign up at [together.ai](https://www.together.ai/), generate an API key, and update the relevant Nemotron LLM blocks. The default config uses `nemotron_super_llm` and `nemotron_ultra_llm`; the pattern below shows the provider change for one block:\n",
+    "For example, to use [Together.ai](https://www.together.ai/) with a supported model — sign up at [together.ai](https://www.together.ai/), generate an API key, and update the relevant Lightning or Ultra LLM block. The pattern below shows the provider change for one block:\n",
     "\n",
     "```yaml\n",
     "llms:\n",
-    "  nemotron_super_llm:\n",
+    "  nemotron_ultra_llm:\n",
     "    _type: openai\n",
-    "    model_name: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16\n",
+    "    model_name: ${TOGETHER_MODEL_NAME}\n",
     "    base_url: \"https://api.together.ai/v1\"\n",
     "    api_key: ${TOGETHER_API_KEY}\n",
     "    temperature: 0.1\n",
diff --git a/docs/source/customization/configuration-reference.md b/docs/source/customization/configuration-reference.md
index c40df97ad..19a1bef99 100644
--- a/docs/source/customization/configuration-reference.md
+++ b/docs/source/customization/configuration-reference.md
@@ -119,7 +119,7 @@ llms:
 | `top_p` | `float` | `None` | Nucleus sampling threshold. When `None`, the API uses its server-side default. |
 | `max_tokens` | `int` | `300` | Maximum tokens in the response. Set higher values (for example, `16384` or `128000`) for research agents. |
 | `num_retries` | `int` | `5` | Number of retry attempts on API failure. |
-| `parallel_tool_calls` | `bool` | Provider default | Whether the provider can emit parallel tool calls. |
+| `parallel_tool_calls` | `bool` | Provider default | Whether the provider can emit parallel tool calls. The default intent and shallow profiles set this to `false`. |
 | `chat_template_kwargs` | `object` | -- | Extra arguments passed to the chat template. Use `enable_thinking: true` to activate the model's chain-of-thought reasoning. |
 
 ### Common LLM Configurations
@@ -128,8 +128,8 @@ Different agents benefit from different parameter profiles:
 
 | Role | Temperature | Top-p | Max Tokens | Notes |
 |------|------------|-------|------------|-------|
-| Intent classifier (Super) | `0.5` | `0.9` | `4096` | Existing classification profile; thinking enabled |
-| Shallow researcher (Super) | `0.7` | `0.7` | `65536` | Existing research profile; thinking enabled |
+| Intent classifier (Nemotron 3.5 Lightning) | `0.1` | `0.9` | `1024` | Short deterministic classification; thinking disabled |
+| Shallow researcher (Nemotron 3.5 Lightning) | `0.2` | `0.7` | `8192` | Tool-calling profile; parallel tool calls disabled and thinking enabled |
 | Deep research roles (Ultra) | `0.2` | `0.7` | `16384` | Source routing, orchestration, planning, and research |
 | Deep research writer (Ultra) | `0.2` | `0.7` | `32768` | Larger report-writing budget |
 | Summary LLM (Gemma) | `0.1` | -- | `100` | Conservative, short document summaries |
@@ -379,7 +379,7 @@ Classifies user queries as meta (conversational) or research, and determines res
 functions:
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     tools:
       - web_search_tool
       - paper_search_tool
@@ -588,25 +588,29 @@ general:
 
 # LLM definitions
 llms:
-  super_intent_llm:                     # Used by intent classifier
+  lightning_intent_llm:                # Used by intent classifier
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  super_agent_llm:                      # Used by shallow researcher
+  lightning_agent_llm:                 # Used by shallow researcher
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -653,7 +657,7 @@ functions:
 
   intent_classifier:                   # Classifies queries, routes depth
     _type: intent_classifier
-    llm: super_intent_llm
+    llm: lightning_intent_llm
     tools:
       - web_search_tool
       - paper_search_tool
@@ -668,7 +672,7 @@ functions:
 
   shallow_research_agent:              # Fast single-pass research
     _type: shallow_research_agent
-    llm: super_agent_llm
+    llm: lightning_agent_llm
     tools:
       - web_search_tool
     max_llm_turns: 10
@@ -706,7 +710,7 @@ only the additional sections you need.
 | `configs/config_web_azure_ai_search.yml` | Web API | Azure AI Search knowledge retrieval and web search |
 | `configs/config_web_frag.yml` | Web API / Helm base | Foundational RAG plus Tavily. Requires separately deployed RAG query and ingestion services. Paper search is commented out. |
 | `configs/config_web_opensearch.yml` | Web API | Built-in OpenSearch knowledge backend plus Tavily. Supports unauthenticated or basic self-hosted OpenSearch and SigV4 (`es` or `aoss`); infrastructure and credentials are deployment opt-ins. |
-| `configs/config_frontier_models.yml` | Web API | Shipped LlamaIndex frontier profile: Nemotron 3 Super for intent/shallow, GPT-5.6 Sol for clarification/orchestration/planning/writing, GPT-5.6 Luna for source routing/research, and Gemma 4 for summaries. Requires `NVIDIA_API_KEY`, `OPENAI_API_KEY`, and `TAVILY_API_KEY` for the enabled Tavily tools; the commented paper-search opt-in requires `SERPER_API_KEY` when enabled. Validate the complete workflow against the configured provider endpoints before deployment. |
+| `configs/config_frontier_models.yml` | Web API | Shipped LlamaIndex frontier profile: Nemotron 3.5 Lightning for intent/shallow, GPT-5.6 Sol for clarification/orchestration/planning/writing, GPT-5.6 Luna for source routing/research, and Gemma 4 for summaries. Requires `NVIDIA_API_KEY`, `OPENAI_API_KEY`, and `TAVILY_API_KEY` for the enabled Tavily tools; the commented paper-search opt-in requires `SERPER_API_KEY` when enabled. Validate the complete workflow against the configured provider endpoints before deployment. |
 | `configs/config_web_default_guardrails.yml` | Web API | LlamaIndex with workflow Guardrails attached explicitly, shallow-agent Guardrails dynamically attached through `workflow_functions`, and async deep-agent Guardrails applied by the AI-Q runner from the same target configuration. |
 | `configs/config_web_frag_mcp_auth.yml` | Web API | Foundational RAG plus a protected per-user OAuth MCP source example. Requires a real protected MCP endpoint and shared token-store configuration; it is not a zero-config default. |
 | `configs/config_domain_routing_and_skills.yml` | Direct deep-research workflow | Automatic domain routing, Tavily, DuckDuckGo news, Polymarket, LlamaIndex, enabled Serper paper search, built-in skills, and a Modal sandbox. Requires the corresponding service credentials and Modal setup. |
diff --git a/docs/source/customization/mcp-tools.md b/docs/source/customization/mcp-tools.md
index d2597d6d1..97a50a968 100644
--- a/docs/source/customization/mcp-tools.md
+++ b/docs/source/customization/mcp-tools.md
@@ -114,7 +114,7 @@ Use `exclude_tools` to specialize individual agents:
 functions:
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     exclude_tools:
       - mcp_financial_tools__expensive_long_running_tool
 ```
diff --git a/docs/source/customization/swapping-models.md b/docs/source/customization/swapping-models.md
index a043e0e57..675192d3f 100644
--- a/docs/source/customization/swapping-models.md
+++ b/docs/source/customization/swapping-models.md
@@ -12,8 +12,8 @@ AI-Q 2.2 ships these exact model assignments and parameters:
 
 | Configuration | Intent and shallow roles | Clarification and deep-research roles | Optional summary role |
 | --- | --- | --- | --- |
-| `configs/config_cli_default.yml`, `configs/config_web_default_llamaindex.yml` | `nvidia/nemotron-3-super-120b-a12b` | `nvidia/nemotron-3-ultra-550b-a55b` for clarification, orchestration, source routing, research, planning, and writing | `google/gemma-4-31b-it` in the web profile |
-| `configs/config_frontier_models.yml` | `nvidia/nemotron-3-super-120b-a12b` | `gpt-5.6-sol` for clarification, orchestration, planning, and writing; `gpt-5.6-luna` for source routing and research | `google/gemma-4-31b-it` |
+| `configs/config_cli_default.yml`, `configs/config_web_default_llamaindex.yml` | `nvidia/nemotron-3.5-lightning-30b-a3b` | `nvidia/nemotron-3-ultra-550b-a55b` for clarification, orchestration, source routing, research, planning, and writing | `google/gemma-4-31b-it` in the web profile |
+| `configs/config_frontier_models.yml` | `nvidia/nemotron-3.5-lightning-30b-a3b` | `gpt-5.6-sol` for clarification, orchestration, planning, and writing; `gpt-5.6-luna` for source routing and research | `google/gemma-4-31b-it` |
 
 The checked-in files define the documented compatibility boundary; they are not a substitute for an end-to-end run
 against your provider endpoints and credentials. Changing a model, endpoint, role assignment, prompt, or inference
@@ -26,14 +26,18 @@ evaluated in that exact configuration.
 
 ```yaml
 llms:
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
     max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
+    chat_template_kwargs:
+      enable_thinking: true
 ```
 
 **Example: NIM with thinking (for example, for deep research)**
@@ -52,7 +56,7 @@ llms:
       enable_thinking: true
 ```
 
-**Model roles:** The workflow maps LLMs to roles (orchestrator, researcher, planner, etc.) through the `LLMProvider`. In YAML you assign which named LLM each agent uses (for example, `orchestrator_llm: nemotron_ultra_llm`, `llm: nemotron_super_llm`). Use different keys in `llms` and point agents at them to swap models per role.
+**Model roles:** The workflow maps LLMs to roles (orchestrator, researcher, planner, etc.) through the `LLMProvider`. In YAML you assign which named LLM each agent uses (for example, `orchestrator_llm: nemotron_ultra_llm`, `llm: nemotron_lightning_agent_llm`). Use different keys in `llms` and point agents at them to swap models per role.
 
 ## Using Downloadable NIMs (Self-Hosted)
 
@@ -88,7 +92,7 @@ llms:
 ```
 
 ```{note}
-**Hosted Endpoint Availability:** The default profiles retain Nemotron 3 Super for intent and shallow research, and use Nemotron 3 Ultra for clarification and every deep-research role. Shared hosted endpoints can have limited availability during high demand (HTTP 429 or 503 responses). For production deployments requiring consistent throughput, refer to the [self-hosting guidance](../resources/troubleshooting.md#nemotron-hosted-endpoint-availability).
+**Hosted Endpoint Availability:** The default profiles use Nemotron 3.5 Lightning for intent and shallow research, and Nemotron 3 Ultra for clarification and every deep-research role. Shared hosted endpoints can have limited availability during high demand (HTTP 429 or 503 responses). For production deployments requiring consistent throughput, refer to the [self-hosting guidance](../resources/troubleshooting.md#nemotron-hosted-endpoint-availability).
 ```
 
 You can mix hosted and local NIMs in the same config -- for example, use a hosted endpoint for shallow research and a local downloadable Ultra NIM for deep research:
@@ -97,11 +101,15 @@ You can mix hosted and local NIMs in the same config -- for example, use a hoste
 llms:
   hosted_shallow_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
     api_key: ${NVIDIA_API_KEY}
-    temperature: 0.7
-    max_tokens: 65536
+    temperature: 0.2
+    top_p: 0.7
+    max_tokens: 8192
+    parallel_tool_calls: false
+    chat_template_kwargs:
+      enable_thinking: true
 
   local_ultra_llm:
     _type: nim
diff --git a/docs/source/customization/tools-and-sources.md b/docs/source/customization/tools-and-sources.md
index e0412b604..f69b86848 100644
--- a/docs/source/customization/tools-and-sources.md
+++ b/docs/source/customization/tools-and-sources.md
@@ -137,7 +137,7 @@ functions:
   # Agents with no tools list inherit all registry tools
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
 
   clarifier_agent:
     _type: clarifier_agent
diff --git a/docs/source/examples/full-pipeline-llamaindex.md b/docs/source/examples/full-pipeline-llamaindex.md
index d1e940fdf..c639b2a9c 100644
--- a/docs/source/examples/full-pipeline-llamaindex.md
+++ b/docs/source/examples/full-pipeline-llamaindex.md
@@ -43,25 +43,29 @@ general:
 # LLMs
 # ===========================================================================
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.1
-    top_p: 0.3
-    max_tokens: 16384
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
+    top_p: 0.7
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -137,7 +141,7 @@ functions:
 
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     verbose: true
     tools:
       - web_search_tool
@@ -156,7 +160,7 @@ functions:
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     verbose: true
     tools:
       - web_search_tool
diff --git a/docs/source/examples/full-pipeline-web.md b/docs/source/examples/full-pipeline-web.md
index 3e5acdf82..f678552f1 100644
--- a/docs/source/examples/full-pipeline-web.md
+++ b/docs/source/examples/full-pipeline-web.md
@@ -61,34 +61,35 @@ general:
 # LLMs
 # ===========================================================================
 # Role-specific LLM configurations:
-# - Super for intent classification and shallow research
+# - Nemotron 3.5 Lightning for intent classification and shallow research
 # - Ultra for clarification and every deep-research role
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.5    # Moderate: needs to reason about intent
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
     top_p: 0.9
-    max_tokens: 4096
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
-      enable_thinking: true
+      enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.1    # Low: factual research output
-    top_p: 0.3
-    max_tokens: 16384
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
+    top_p: 0.7
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
-# ===========================================================================
-# Functions (tools and agents)
-# ===========================================================================
   nemotron_ultra_llm:
     _type: nim
     model_name: nvidia/nemotron-3-ultra-550b-a55b
@@ -156,7 +157,7 @@ functions:
   # Has access to tools for context-aware routing decisions.
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
     tools:
       - web_search_tool
       - paper_search_tool
@@ -183,7 +184,7 @@ functions:
   # Single-turn ReAct agent for quick queries.
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     tools:
       - web_search_tool
       - knowledge_search
diff --git a/docs/source/examples/hybrid-frontier-model.md b/docs/source/examples/hybrid-frontier-model.md
index e416165c8..20a4c3ca7 100644
--- a/docs/source/examples/hybrid-frontier-model.md
+++ b/docs/source/examples/hybrid-frontier-model.md
@@ -12,7 +12,7 @@ shallow research with GPT Sol/Luna for clarification and deep research.
 
 | Role | Model |
 | --- | --- |
-| Intent classification and shallow research | `nvidia/nemotron-3-super-120b-a12b` |
+| Intent classification and shallow research | `nvidia/nemotron-3.5-lightning-30b-a3b` |
 | Clarification, orchestration, and planning | `gpt-5.6-sol` |
 | Source routing and research | `gpt-5.6-luna` |
 | Report writing | `gpt-5.6-sol` with the writer token budget from the checked-in config |
diff --git a/docs/source/examples/minimal-shallow-only.md b/docs/source/examples/minimal-shallow-only.md
index 6f80af2c6..0656caeff 100644
--- a/docs/source/examples/minimal-shallow-only.md
+++ b/docs/source/examples/minimal-shallow-only.md
@@ -25,18 +25,19 @@ general:
 # ---------------------------------------------------------------------------
 # LLMs
 # ---------------------------------------------------------------------------
-# A single role-specific Super LLM is sufficient for shallow research. The NIM type connects
+# A single role-specific Lightning LLM is sufficient for shallow research. The NIM type connects
 # to NVIDIA API Catalog.
 llms:
   research_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
     api_key: ${NVIDIA_API_KEY}
-    temperature: 0.7
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536      # Max output length per LLM call
+    max_tokens: 8192       # Max output length per LLM call
     num_retries: 5         # Retry on transient API failures
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
diff --git a/docs/source/extending/adding-a-tool.md b/docs/source/extending/adding-a-tool.md
index d548d0879..9b25fd3a0 100644
--- a/docs/source/extending/adding-a-tool.md
+++ b/docs/source/extending/adding-a-tool.md
@@ -279,7 +279,7 @@ Reference your tool in any workflow configuration:
 llms:
   research_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-ultra-550b-a55b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
 
 functions:
   my_search:
diff --git a/docs/source/get-started/installation.md b/docs/source/get-started/installation.md
index 8acf44784..19152ccd5 100644
--- a/docs/source/get-started/installation.md
+++ b/docs/source/get-started/installation.md
@@ -25,7 +25,7 @@ When using [NVIDIA API Catalog](https://build.nvidia.com/) (the default), infere
 
 | Component | Default Model | Self-Hosted Hardware Reference |
 |-----------|---------------|-------------------------------|
-| LLM (intent classifier, shallow researcher) | `nvidia/nemotron-3-super-120b-a12b` | [Nemotron 3 Super](https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b) |
+| LLM (intent classifier, shallow researcher) | `nvidia/nemotron-3.5-lightning-30b-a3b` | [Nemotron 3.5 Lightning](https://build.nvidia.com/nvidia/nemotron-3.5-lightning-30b-a3b/modelcard) |
 | LLM (clarifier and all deep-research roles) | `nvidia/nemotron-3-ultra-550b-a55b` | [Nemotron 3 Ultra](https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b) |
 | Document summary (optional) | `google/gemma-4-31b-it` | [Gemma 4 31B IT](https://build.nvidia.com/google/gemma-4-31b-it) |
 | Text embedding | `nvidia/nemotron-3-embed-1b` | [NeMo Retriever embedding support matrix](https://docs.nvidia.com/nim/nemo-retriever/text-embedding/latest/support-matrix.html) |
diff --git a/docs/source/resources/troubleshooting.md b/docs/source/resources/troubleshooting.md
index 5afcc2cda..438d89bf1 100644
--- a/docs/source/resources/troubleshooting.md
+++ b/docs/source/resources/troubleshooting.md
@@ -44,19 +44,18 @@ Common issues and solutions for the AI-Q blueprint.
 
 ## Nemotron Hosted Endpoint Availability
 
-Nemotron 3 Super (`nvidia/nemotron-3-super-120b-a12b`) and Nemotron 3 Ultra (`nvidia/nemotron-3-ultra-550b-a55b`) are compatible and tested with AIQ, but their NVIDIA-hosted endpoints can have limited availability during high demand. During peak periods you may observe:
+Nemotron 3.5 Lightning (`nvidia/nemotron-3.5-lightning-30b-a3b`) and Nemotron 3 Ultra (`nvidia/nemotron-3-ultra-550b-a55b`) are compatible and tested with AIQ, but their NVIDIA-hosted endpoints can have limited availability during high demand. During peak periods you may observe:
 
 - Elevated latency or timeouts on LLM inference calls
 - HTTP 429 (rate-limited) or 503 (service unavailable) responses from the Build API
 - Degraded agent workflow performance due to upstream model availability
 
-**Default Configuration:** The default configs retain Nemotron 3 Super for intent classification and shallow research, and use Nemotron 3 Ultra for clarification and all deep-research roles. If either hosted endpoint is saturated, retry after a short delay, reduce concurrency, or self-host a downloadable model for consistent throughput.
+**Default Configuration:** The default configs use Nemotron 3.5 Lightning for intent classification and shallow research, and Nemotron 3 Ultra for clarification and all deep-research roles. If a hosted endpoint is saturated, retry after a short delay, reduce concurrency, or self-host a downloadable model for consistent throughput.
 
 ### Recommended Mitigation: Self-Host the Affected Model
 
 For production and staging deployments that require consistent throughput and low-latency inference, self-host a downloadable NVIDIA NIM rather than relying on shared endpoints. Preview endpoint availability and downloadable NIM availability do not necessarily move in lockstep; verify the current model card before choosing an image.
 
-- [Self-host Nemotron 3 Super 120B A12B](https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b/deploy?nim=self-hosted) for the default intent and shallow-research roles
 - [Self-host Nemotron 3 Ultra 550B A55B](https://build.nvidia.com/nvidia/nemotron-3-ultra-550b-a55b?nim=self-hosted) for the default clarification and deep-research roles
 
 Once your self-hosted endpoint is running, update the corresponding `base_url` in your config to point at it. AIQ's configuration validator currently requires `NVIDIA_API_KEY` for every `_type: nim` profile, even when a local NIM does not enforce client authentication. Set a non-secret placeholder for the local deployment before starting AIQ:
diff --git a/frontends/aiq_api/src/aiq_api/jobs/callbacks.py b/frontends/aiq_api/src/aiq_api/jobs/callbacks.py
index 1ecd320fe..789a8ca65 100644
--- a/frontends/aiq_api/src/aiq_api/jobs/callbacks.py
+++ b/frontends/aiq_api/src/aiq_api/jobs/callbacks.py
@@ -42,6 +42,7 @@
 from pydantic import ConfigDict
 from pydantic import Field
 
+from aiq_agent.common.callbacks import SUPPRESS_OUTPUT_ARTIFACT_TAG
 from aiq_agent.common.citation_verification import extract_http_urls
 from aiq_agent.common.citation_verification import get_session_registry
 
@@ -740,7 +741,7 @@ def on_llm_start(self, serialized: dict, prompts: list, **kwargs) -> None:
 
     def on_llm_new_token(self, token: str, **kwargs) -> None:
         """Emit a streaming llm.chunk event for each non-empty token."""
-        if token:
+        if token and SUPPRESS_OUTPUT_ARTIFACT_TAG not in (kwargs.get("tags") or []):
             self._emit(
                 IntermediateStepEvent(
                     category=EventCategory.LLM,
@@ -784,6 +785,7 @@ def on_llm_end(self, response, **kwargs) -> None:
             and len(content) >= self.OUTPUT_MIN_LENGTH
             and not has_tool_calls
             and not self._contains_tool_call_syntax(content)
+            and SUPPRESS_OUTPUT_ARTIFACT_TAG not in (kwargs.get("tags") or [])
         ):
             output_category = self._get_output_category(agent_info)
             self._emit_artifact(
diff --git a/frontends/benchmarks/freshqa/configs/config_full_workflow.yml b/frontends/benchmarks/freshqa/configs/config_full_workflow.yml
index af1cbc247..33a629c49 100644
--- a/frontends/benchmarks/freshqa/configs/config_full_workflow.yml
+++ b/frontends/benchmarks/freshqa/configs/config_full_workflow.yml
@@ -9,25 +9,29 @@ general:
         level: INFO
 
 llms:
-  nemotron_llm_intent:
+  nemotron_lightning_intent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
-    top_p: 0.7
-    max_tokens: 8192
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.1
+    top_p: 0.9
+    max_tokens: 1024
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: false
 
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -63,17 +67,26 @@ llms:
     temperature: 0.1
 
 functions:
+  data_sources:
+    _type: data_source_registry
+    sources:
+      - id: web_search
+        name: "Web Search"
+        description: "Search the web for real-time information."
+        tools:
+          - web_search_tool
+
   web_search_tool:
-    _type: tavily_internet_search
+    _type: tavily_web_search
     max_results: 5
 
   intent_classifier:
     _type: intent_classifier
-    llm: nemotron_llm_intent
+    llm: nemotron_lightning_intent_llm
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     tools:
       - web_search_tool
     max_llm_turns: 10
diff --git a/frontends/benchmarks/freshqa/configs/config_shallow_research_only.yml b/frontends/benchmarks/freshqa/configs/config_shallow_research_only.yml
index b2900c1b4..9db321278 100644
--- a/frontends/benchmarks/freshqa/configs/config_shallow_research_only.yml
+++ b/frontends/benchmarks/freshqa/configs/config_shallow_research_only.yml
@@ -1,5 +1,5 @@
 # FreshQA shallow-research-only
-# Nemotron Ultra on NVIDIA API Catalog; add judge (OpenAI or Gemini) for evaluator.
+# Nemotron 3.5 Lightning on NVIDIA API Catalog; add judge (OpenAI or Gemini) for evaluator.
 
 general:
   telemetry:
@@ -13,14 +13,16 @@ general:
         project: "AIQ_v2_freshqa"
 
 llms:
-  nemotron_super_llm:
+  nemotron_lightning_agent_llm:
     _type: nim
-    model_name: nvidia/nemotron-3-super-120b-a12b
+    model_name: nvidia/nemotron-3.5-lightning-30b-a3b
     base_url: "https://integrate.api.nvidia.com/v1"
-    temperature: 0.7
+    api_key: ${NVIDIA_API_KEY}
+    temperature: 0.2
     top_p: 0.7
-    max_tokens: 65536
+    max_tokens: 8192
     num_retries: 5
+    parallel_tool_calls: false
     chat_template_kwargs:
       enable_thinking: true
 
@@ -31,13 +33,22 @@ llms:
     temperature: 0.1
 
 functions:
+  data_sources:
+    _type: data_source_registry
+    sources:
+      - id: web_search
+        name: "Web Search"
+        description: "Search the web for real-time information."
+        tools:
+          - web_search_tool
+
   web_search_tool:
-    _type: tavily_internet_search
+    _type: tavily_web_search
     max_results: 5
 
   shallow_research_agent:
     _type: shallow_research_agent
-    llm: nemotron_super_llm
+    llm: nemotron_lightning_agent_llm
     max_llm_turns: 20
     tools:
       - web_search_tool
diff --git a/frontends/ui/package-lock.json b/frontends/ui/package-lock.json
index 72afdb34c..b3064174b 100644
--- a/frontends/ui/package-lock.json
+++ b/frontends/ui/package-lock.json
@@ -6882,9 +6882,9 @@
       "license": "MIT"
     },
     "node_modules/dompurify": {
-      "version": "3.4.12",
-      "resolved": "https://registry.npmjs.org/dompurify/-/dompurify-3.4.12.tgz",
-      "integrity": "sha512-zQvGet8Z2sWbQhCmfFz/T5QWH2oBmjnqK3qvOjaqaNLrLEF912WamU+ohnTp0TCep/MFVHpdJuCZEdFOdTnEFg==",
+      "version": "3.4.13",
+      "resolved": "https://registry.npmjs.org/dompurify/-/dompurify-3.4.13.tgz",
+      "integrity": "sha512-2vmYIoqjze2d+kakP8S/nS5shfsl587kzwEjcGlTdiksUVgFHnFCsLYDVj/JNqJVOQZGSYBTmuycv0PodwmnMQ==",
       "license": "(MPL-2.0 OR Apache-2.0)",
       "optionalDependencies": {
         "@types/trusted-types": "^2.0.7"
@@ -10863,9 +10863,9 @@
       }
     },
     "node_modules/nanoid": {
-      "version": "3.3.16",
-      "resolved": "https://registry.npmjs.org/nanoid/-/nanoid-3.3.16.tgz",
-      "integrity": "sha512-bzlKTyNJ7+LdGIIwy8ijFpIqEQIvafahV7eYykJ8Cvh42EdJeODoJ6gUJXpQJvej1BddH8OqTXZNE/KfbWAu8Q==",
+      "version": "3.3.18",
+      "resolved": "https://registry.npmjs.org/nanoid/-/nanoid-3.3.18.tgz",
+      "integrity": "sha512-DTg4MJbGMWkfi6VZFdNt2/caMbQy4Ou+Op/hJQvGEWcnVfoA1QA+xzRKAzw9jD6+GVOOeYr/mIcuDSdug6F6+w==",
       "funding": [
         {
           "type": "github",
diff --git a/mcp/tests/test_preclassification_integration.py b/mcp/tests/test_preclassification_integration.py
index 315b5f50e..1cf2b4292 100644
--- a/mcp/tests/test_preclassification_integration.py
+++ b/mcp/tests/test_preclassification_integration.py
@@ -27,7 +27,7 @@
 from nat.builder.workflow_builder import WorkflowBuilder
 
 _REPO_ROOT = Path(__file__).resolve().parents[2]
-_INTENT_LLM_NAME = "nemotron_llm_intent"
+_INTENT_LLM_NAME = "nemotron_lightning_intent_llm"
 
 
 class _CountingIntentLLM:
diff --git a/src/aiq_agent/agents/shallow_researcher/agent.py b/src/aiq_agent/agents/shallow_researcher/agent.py
index 92b117d70..f1fa7ca4a 100644
--- a/src/aiq_agent/agents/shallow_researcher/agent.py
+++ b/src/aiq_agent/agents/shallow_researcher/agent.py
@@ -38,6 +38,7 @@
 from aiq_agent.common import get_source_id_for_tool
 from aiq_agent.common import load_prompt
 from aiq_agent.common import render_prompt_template
+from aiq_agent.common.callbacks import SUPPRESS_OUTPUT_ARTIFACT_TAG
 from aiq_agent.common.citation_verification import EmptySourceRegistryError
 from aiq_agent.common.citation_verification import SourceEntry
 from aiq_agent.common.citation_verification import SourceRegistry
@@ -190,6 +191,8 @@ def _get_llm(self) -> BaseChatModel:
     def _build_graph(self) -> CompiledStateGraph:
         """Build the LangGraph StateGraph."""
 
+        source_tool_names = {tool.name for tool in self.tools}
+
         async def agent_node(state: ShallowResearchAgentState) -> dict[str, Any]:
             """Execute the agent with parallel call tracking and context anchoring."""
             messages = state.messages
@@ -244,9 +247,30 @@ async def agent_node(state: ShallowResearchAgentState) -> dict[str, Any]:
                     return {"messages": [response], "tool_iterations": iterations}
 
                 llm = self._get_llm()
-                llm_with_tools = llm.bind_tools(self.tools, parallel_tool_calls=True) if self.tools else llm
+                llm_with_tools = llm.bind_tools(self.tools) if self.tools else llm
                 full_messages = [system_message] + processed_history
-                response = await llm_with_tools.ainvoke(full_messages)
+                pre_evidence_config = {"tags": [SUPPRESS_OUTPUT_ARTIFACT_TAG]} if iterations == 0 else None
+                response = await llm_with_tools.ainvoke(full_messages, config=pre_evidence_config)
+
+                if self.tools and iterations == 0 and not getattr(response, "tool_calls", None):
+                    logger.warning("Shallow researcher returned an answer before collecting evidence; retrying once")
+                    tool_required = HumanMessage(
+                        content=(
+                            "Research is required before answering. Call exactly one available research tool now. "
+                            "Do not provide a final answer until the tool result is available."
+                        )
+                    )
+                    retry_llm = llm.bind_tools(self.tools, parallel_tool_calls=False)
+                    response = await retry_llm.ainvoke(
+                        full_messages + [response, tool_required],
+                        config=pre_evidence_config,
+                    )
+                    retry_tool_calls = getattr(response, "tool_calls", None) or []
+                    if len(retry_tool_calls) != 1 or retry_tool_calls[0].get("name") not in source_tool_names:
+                        raise RuntimeError(
+                            "shallow_research_tool_required: model did not call exactly one allowed research tool "
+                            "after one retry"
+                        )
 
                 new_iterations = iterations
                 if hasattr(response, "tool_calls") and response.tool_calls:
@@ -275,8 +299,6 @@ async def agent_node(state: ShallowResearchAgentState) -> dict[str, Any]:
         # data_source_registry then decides which of those are configured
         # data sources. Having both gates keeps behavior consistent across
         # agents and safe even if the global registry is ever polluted.
-        source_tool_names = {t.name for t in self.tools}
-
         async def tool_node_with_source_capture(state: ShallowResearchAgentState) -> dict[str, Any]:
             """Execute tools and capture source URLs/citations for verification.
 
diff --git a/src/aiq_agent/agents/shallow_researcher/prompts/researcher.j2 b/src/aiq_agent/agents/shallow_researcher/prompts/researcher.j2
index 72c10cd53..eba305767 100644
--- a/src/aiq_agent/agents/shallow_researcher/prompts/researcher.j2
+++ b/src/aiq_agent/agents/shallow_researcher/prompts/researcher.j2
@@ -1,5 +1,13 @@
 You are a Shallow Research Agent. Your role is to provide rapid, citation-backed answers using available tools.
 
+## MANDATORY: Research Before Answering
+When at least one research tool is available, your first response MUST be a tool call. Never answer directly from memory.
+
+- This rule applies even when the question looks simple, familiar, or based on a false premise. Search to verify the answer or premise.
+- Do not produce a final answer until at least one research tool call has completed.
+- If the first search is empty or irrelevant, rewrite the query and search once more before synthesizing.
+- After receiving useful results, answer promptly. Do not keep searching when one call provides enough evidence.
+
 ## CRITICAL: Source Hierarchy
 Prioritize sources based on the query intent:
 
diff --git a/src/aiq_agent/common/callbacks.py b/src/aiq_agent/common/callbacks.py
index f3031f962..f68e90abb 100644
--- a/src/aiq_agent/common/callbacks.py
+++ b/src/aiq_agent/common/callbacks.py
@@ -26,6 +26,9 @@
 logger = logging.getLogger(__name__)
 
 
+SUPPRESS_OUTPUT_ARTIFACT_TAG = "aiq:suppress-output-artifact"
+
+
 YELLOW = "\033[33m"
 CYAN = "\033[36m"
 GREEN = "\033[32m"
diff --git a/tests/aiq_agent/agents/shallow_researcher/test_agent.py b/tests/aiq_agent/agents/shallow_researcher/test_agent.py
index 6144a3e1d..00e9d96f6 100644
--- a/tests/aiq_agent/agents/shallow_researcher/test_agent.py
+++ b/tests/aiq_agent/agents/shallow_researcher/test_agent.py
@@ -17,6 +17,7 @@
 
 from unittest.mock import AsyncMock
 from unittest.mock import MagicMock
+from unittest.mock import call
 from unittest.mock import patch
 
 import pytest
@@ -29,6 +30,7 @@
 from aiq_agent.agents.shallow_researcher.models import ShallowResearchAgentState
 from aiq_agent.common import LLMProvider
 from aiq_agent.common import LLMRole
+from aiq_agent.common.callbacks import SUPPRESS_OUTPUT_ARTIFACT_TAG
 from aiq_agent.common.citation_verification import EmptySourceRegistryError
 from aiq_agent.common.citation_verification import EmptySourceRegistryReason
 from aiq_agent.common.citation_verification import SourceEntry
@@ -200,7 +202,7 @@ async def test_run_basic_query(self, mock_llm_provider, mock_llm, real_tool):
 
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[real_tool],
+            tools=[],
         )
 
         state = ShallowResearchAgentState(messages=[HumanMessage(content="What is CUDA?")])
@@ -219,7 +221,7 @@ async def test_run_with_callbacks(self, mock_llm_provider, mock_llm, real_tool):
         mock_callback = MagicMock()
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[real_tool],
+            tools=[],
             callbacks=[mock_callback],
         )
 
@@ -239,7 +241,7 @@ async def test_run_with_user_info(self, mock_llm_provider, mock_llm, real_tool):
         custom_prompt = "You are an assistant. User: {{ user_info }}."
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[real_tool],
+            tools=[],
             system_prompt=custom_prompt,
         )
 
@@ -260,7 +262,7 @@ async def test_run_with_tools_info_in_state(self, mock_llm_provider, mock_llm, r
 
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[real_tool],
+            tools=[],
         )
 
         custom_tools_info = [
@@ -299,6 +301,77 @@ def test_default_prompt_requires_tool_result_references(self, mock_llm_provider,
         assert "exact tool name" in agent.system_prompt
         assert "- [1] mcp_time__get_current_time" in agent.system_prompt
 
+    @pytest.mark.asyncio
+    async def test_initial_answer_without_tool_call_is_retried(self, mock_llm_provider, mock_llm, real_tool):
+        """An initial memory-only answer is retried and replaced by a tool call."""
+        initial_answer = AIMessage(content="Memory-only answer")
+        tool_call = AIMessage(
+            content="",
+            tool_calls=[{"name": "web_search_tool", "args": {"query": "CUDA"}, "id": "retry-tool"}],
+        )
+        final_answer = AIMessage(content="Evidence-backed answer")
+        mock_llm.ainvoke = AsyncMock(side_effect=[initial_answer, tool_call, final_answer])
+
+        agent = ShallowResearcherAgent(llm_provider=mock_llm_provider, tools=[real_tool])
+        result = await agent.run(ShallowResearchAgentState(messages=[HumanMessage(content="What is CUDA?")]))
+
+        assert result.messages[-1].content == "Evidence-backed answer"
+        assert mock_llm.ainvoke.await_count == 3
+        assert mock_llm.bind_tools.call_args_list[:2] == [
+            call([real_tool]),
+            call([real_tool], parallel_tool_calls=False),
+        ]
+        for invocation in mock_llm.ainvoke.await_args_list[:2]:
+            assert invocation.kwargs["config"] == {"tags": [SUPPRESS_OUTPUT_ARTIFACT_TAG]}
+
+    @pytest.mark.asyncio
+    async def test_repeated_answer_without_tool_call_fails_closed(self, mock_llm_provider, mock_llm, real_tool):
+        """A model that ignores the bounded tool-use retry cannot synthesize an answer."""
+        mock_llm.ainvoke = AsyncMock(
+            side_effect=[AIMessage(content="First answer"), AIMessage(content="Second answer")]
+        )
+        agent = ShallowResearcherAgent(llm_provider=mock_llm_provider, tools=[real_tool])
+
+        with pytest.raises(RuntimeError, match="shallow_research_tool_required"):
+            await agent.run(ShallowResearchAgentState(messages=[HumanMessage(content="What is CUDA?")]))
+
+    @pytest.mark.asyncio
+    async def test_retry_with_multiple_tool_calls_fails_closed(self, mock_llm_provider, mock_llm, real_tool):
+        """The bounded retry cannot schedule multiple tools or exceed its one-call contract."""
+        multiple_tool_calls = AIMessage(
+            content="",
+            tool_calls=[
+                {"name": "web_search_tool", "args": {"query": "CUDA"}, "id": "retry-tool-1"},
+                {"name": "web_search_tool", "args": {"query": "GPU"}, "id": "retry-tool-2"},
+            ],
+        )
+        mock_llm.ainvoke = AsyncMock(side_effect=[AIMessage(content="Memory-only answer"), multiple_tool_calls])
+        agent = ShallowResearcherAgent(llm_provider=mock_llm_provider, tools=[real_tool])
+
+        with pytest.raises(RuntimeError, match="exactly one allowed research tool"):
+            await agent.run(ShallowResearchAgentState(messages=[HumanMessage(content="What is CUDA?")]))
+
+        assert mock_llm.ainvoke.await_count == 2
+        assert mock_llm.bind_tools.call_args_list == [
+            call([real_tool]),
+            call([real_tool], parallel_tool_calls=False),
+        ]
+
+    @pytest.mark.asyncio
+    async def test_retry_with_unknown_tool_fails_closed(self, mock_llm_provider, mock_llm, real_tool):
+        """The bounded retry cannot schedule a tool outside the agent's allowlist."""
+        unknown_tool_call = AIMessage(
+            content="",
+            tool_calls=[{"name": "unavailable_tool", "args": {}, "id": "retry-tool"}],
+        )
+        mock_llm.ainvoke = AsyncMock(side_effect=[AIMessage(content="Memory-only answer"), unknown_tool_call])
+        agent = ShallowResearcherAgent(llm_provider=mock_llm_provider, tools=[real_tool])
+
+        with pytest.raises(RuntimeError, match="exactly one allowed research tool"):
+            await agent.run(ShallowResearchAgentState(messages=[HumanMessage(content="What is CUDA?")]))
+
+        assert mock_llm.ainvoke.await_count == 2
+
     @pytest.mark.asyncio
     async def test_tool_iterations_incremented_on_tool_calls(self, mock_llm_provider, mock_llm, real_tool):
         """Test tool_iterations counter increments when LLM makes tool calls."""
@@ -377,7 +450,7 @@ async def test_run_returns_updated_tool_iterations(self, mock_llm_provider, mock
 
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[real_tool],
+            tools=[],
         )
 
         state = ShallowResearchAgentState(
@@ -868,7 +941,7 @@ async def test_run_uses_session_registry_when_set(self, mock_llm_provider, mock_
 
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[web_search_tool],
+            tools=[],
         )
 
         set_session_registry(session_reg)
@@ -895,7 +968,7 @@ async def test_run_clears_registry_in_standalone_mode(self, mock_llm_provider, m
 
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[web_search_tool],
+            tools=[],
         )
         # Pre-populate the instance registry (simulating stale data)
         agent.source_registry.add(SourceEntry(url="https://stale.example.com"))
@@ -922,7 +995,7 @@ async def test_empty_registry_classification_preserves_sanitized_answer(
         expected_reason,
     ):
         mock_llm.ainvoke = AsyncMock(return_value=AIMessage(content="Draft answer with https://private.example/path"))
-        agent = ShallowResearcherAgent(llm_provider=mock_llm_provider, tools=[web_search_tool])
+        agent = ShallowResearcherAgent(llm_provider=mock_llm_provider, tools=[])
         state = ShallowResearchAgentState(
             messages=[HumanMessage(content="Test")],
             data_sources=data_sources,
@@ -1003,7 +1076,7 @@ async def test_session_registry_does_not_mutate_shared_instance(self, mock_llm_p
 
         agent = ShallowResearcherAgent(
             llm_provider=mock_llm_provider,
-            tools=[web_search_tool],
+            tools=[],
         )
         original_registry = agent.source_registry
 
diff --git a/tests/aiq_agent/jobs/test_runner.py b/tests/aiq_agent/jobs/test_runner.py
index 52566fe30..3a7c441b7 100644
--- a/tests/aiq_agent/jobs/test_runner.py
+++ b/tests/aiq_agent/jobs/test_runner.py
@@ -73,6 +73,7 @@
 
 from aiq_agent.agents.deep_researcher.custom_middleware import FinalReportCommitTracker
 from aiq_agent.auth import Principal
+from aiq_agent.common.callbacks import SUPPRESS_OUTPUT_ARTIFACT_TAG
 from aiq_api.jobs.callbacks import ArtifactType
 from aiq_api.jobs.callbacks import DeepResearchEventCallback
 from aiq_api.jobs.callbacks import EventCategory
@@ -373,6 +374,19 @@ def test_on_chat_model_start_emits_event(self):
         assert call_args["type"] == "llm.start"
         assert call_args["name"] == "gpt-4"
 
+    def test_suppressed_llm_output_keeps_lifecycle_without_publishing_content(self):
+        """Pre-evidence answers retain telemetry but cannot reach output artifacts or chunks."""
+        mock_store = MagicMock()
+        callback = DeepResearchEventCallback(event_store=mock_store)
+        tags = [SUPPRESS_OUTPUT_ARTIFACT_TAG]
+
+        callback.on_llm_new_token("rejected draft", tags=tags)
+        callback._extract_llm_response = MagicMock(return_value=("x" * 200, None, None, False))
+        callback.on_llm_end(MagicMock(), run_id="pre-evidence", tags=tags)
+
+        events = [stored.args[0] for stored in mock_store.store.call_args_list]
+        assert [event["type"] for event in events] == ["llm.end"]
+
 
 class TestSubmitDeepResearchJob:
     """Tests for the submit_deep_research_job function."""
diff --git a/tests/aiq_agent/test_default_model_profiles.py b/tests/aiq_agent/test_default_model_profiles.py
index 3372111bb..39baa4f50 100644
--- a/tests/aiq_agent/test_default_model_profiles.py
+++ b/tests/aiq_agent/test_default_model_profiles.py
@@ -9,7 +9,8 @@
 REPO_ROOT = Path(__file__).resolve().parents[2]
 
 ULTRA_MODEL = "nvidia/nemotron-3-ultra-550b-a55b"
-SUPER_MODEL = "nvidia/nemotron-3-super-120b-a12b"
+LIGHTNING_MODEL = "nvidia/nemotron-3.5-lightning-30b-a3b"
+BUILD_BASE_URL = "https://integrate.api.nvidia.com/v1"
 
 CONFIG_GLOBS = (
     ".agents/skills/aiq-configure-workflow/assets/config-scaffold.yml",
@@ -17,11 +18,10 @@
     "frontends/benchmarks/**/configs/*.yml",
 )
 CONFIG_PATHS = tuple(sorted(path for pattern in CONFIG_GLOBS for path in REPO_ROOT.glob(pattern)))
+FRESHQA_CONFIG_PATHS = tuple(sorted(REPO_ROOT.glob("frontends/benchmarks/freshqa/configs/*.yml")))
 
-# Super remains intentionally pinned for intent and shallow research until the
-# separately gated Nano 3.5 follow-up is available. Every other deprecated
-# reference must be removed by this migration.
-REPLACED_REFERENCES = (
+DEPRECATED_REFERENCES = (
+    "/".join(("nvidia", "nemotron-3-super-120b-a12b")),
     "/".join(("nvidia", "nemotron-3-nano-30b-a3b")),
     "/".join(("nvidia", "nemotron-mini-4b-instruct")),
     "/".join(("nvidia", "llama-nemotron-embed-vl-1b-v2")),
@@ -63,6 +63,22 @@ def _model_for_alias(config: dict, alias: str) -> str:
     return config["llms"][alias]["model_name"]
 
 
+def _thinking_enabled(config: dict, alias: str) -> bool:
+    return bool(config["llms"][alias].get("chat_template_kwargs", {}).get("enable_thinking", False))
+
+
+def _registered_source_tools(config: dict) -> set[str]:
+    functions = config.get("functions", {})
+    registries = (
+        function
+        for function in functions.values()
+        if isinstance(function, dict) and function.get("_type") == "data_source_registry"
+    )
+    return {
+        tool for registry in registries for source in registry.get("sources", []) for tool in source.get("tools", [])
+    }
+
+
 @pytest.mark.parametrize("config_path", CONFIG_PATHS, ids=lambda path: str(path.relative_to(REPO_ROOT)))
 def test_default_profiles_use_role_appropriate_models(config_path: Path):
     config = _load_config(config_path)
@@ -74,9 +90,27 @@ def test_default_profiles_use_role_appropriate_models(config_path: Path):
 
         function_type = function.get("_type")
         if function_type == "intent_classifier":
-            assert _model_for_alias(config, function["llm"]) == SUPER_MODEL
+            alias = function["llm"]
+            assert alias == "nemotron_lightning_intent_llm"
+            assert _model_for_alias(config, alias) == LIGHTNING_MODEL
+            assert config["llms"][alias]["base_url"] == BUILD_BASE_URL
+            assert config["llms"][alias]["api_key"] == "${NVIDIA_API_KEY}"
+            assert config["llms"][alias]["temperature"] == 0.1
+            assert config["llms"][alias]["top_p"] == 0.9
+            assert config["llms"][alias]["max_tokens"] == 1024
+            assert not config["llms"][alias]["parallel_tool_calls"]
+            assert not _thinking_enabled(config, alias)
         elif function_type == "shallow_research_agent":
-            assert _model_for_alias(config, function["llm"]) == SUPER_MODEL
+            alias = function["llm"]
+            assert alias == "nemotron_lightning_agent_llm"
+            assert _model_for_alias(config, alias) == LIGHTNING_MODEL
+            assert config["llms"][alias]["base_url"] == BUILD_BASE_URL
+            assert config["llms"][alias]["api_key"] == "${NVIDIA_API_KEY}"
+            assert config["llms"][alias]["temperature"] == 0.2
+            assert config["llms"][alias]["top_p"] == 0.7
+            assert config["llms"][alias]["max_tokens"] == 8192
+            assert not config["llms"][alias]["parallel_tool_calls"]
+            assert _thinking_enabled(config, alias)
         elif config_path.name != "config_frontier_models.yml" and function_type == "clarifier_agent":
             assert _model_for_alias(config, function["llm"]) == ULTRA_MODEL
         elif config_path.name != "config_frontier_models.yml" and function_type == "deep_research_agent":
@@ -91,7 +125,20 @@ def test_default_profiles_use_role_appropriate_models(config_path: Path):
                 assert _model_for_alias(config, function[role]) == ULTRA_MODEL
 
 
-def test_replaced_model_and_endpoint_references_are_absent():
+@pytest.mark.parametrize("config_path", FRESHQA_CONFIG_PATHS, ids=lambda path: path.name)
+def test_freshqa_research_tools_are_registered_data_sources(config_path: Path):
+    config = _load_config(config_path)
+    source_tools = _registered_source_tools(config)
+
+    for function in config.get("functions", {}).values():
+        if isinstance(function, dict) and function.get("_type") in {
+            "shallow_research_agent",
+            "deep_research_agent",
+        }:
+            assert set(function.get("tools", [])) <= source_tools
+
+
+def test_deprecated_model_and_endpoint_references_are_absent():
     violations: list[str] = []
 
     for path in REPO_ROOT.rglob("*"):
@@ -99,8 +146,8 @@ def test_replaced_model_and_endpoint_references_are_absent():
             continue
 
         text = path.read_text(encoding="utf-8")
-        for reference in REPLACED_REFERENCES:
+        for reference in DEPRECATED_REFERENCES:
             if reference in text:
                 violations.append(f"{path.relative_to(REPO_ROOT)}: {reference}")
 
-    assert not violations, "Replaced references remain:\n" + "\n".join(violations)
+    assert not violations, "Deprecated references remain:\n" + "\n".join(violations)