NVIDIA · cv · Jun 4, 2026 · Jun 4, 2026 · Jun 4, 2026 · Jun 4, 2026
diff --git a/.github/workflows/nightly-e2e.yaml b/.github/workflows/nightly-e2e.yaml
@@ -4,6 +4,8 @@
 # Nightly E2E tests:
 #
 #   cloud-e2e                Cloud inference (NVIDIA Endpoint API) on ubuntu-latest.
+#   agent-turn-latency-e2e   Times one real OpenClaw turn and one real Hermes
+#                            turn through the configured NVIDIA Build model.
 #   messaging-providers-e2e  Validates messaging credential provider/placeholder/L7-proxy chain
 #                            for Telegram + Discord + Slack. Uses fake tokens. Slack additionally
 #                            exercises OpenShell provider-shaped alias resolution (#2085 follow-up).
@@ -94,40 +96,28 @@ on:
         description: >-
           Comma-separated job names to run (empty = all).
           Valid: cloud-e2e, cloud-onboard-e2e, cloud-inference-e2e,
-          skill-agent-e2e, openclaw-skill-cli-e2e, docs-validation-e2e, messaging-providers-e2e,
-          openclaw-slack-pairing-e2e,
-          openclaw-tui-chat-correlation-e2e,
-          issue-4434-tui-unreachable-inference-e2e,
-          issue-3600-gpu-proof-optional-e2e,
-          openclaw-discord-pairing-e2e,
+          agent-turn-latency-e2e, skill-agent-e2e, openclaw-skill-cli-e2e,
+          docs-validation-e2e, messaging-providers-e2e, openclaw-slack-pairing-e2e,
+          openclaw-tui-chat-correlation-e2e, issue-4434-tui-unreachable-inference-e2e,
+          issue-3600-gpu-proof-optional-e2e, openclaw-discord-pairing-e2e,
           issue-4462-scope-upgrade-approval-e2e,
           issue-4462-gateway-pinned-approval-characterization-e2e,
-          messaging-compatible-endpoint-e2e,
-          kimi-inference-compat-e2e,
-          bedrock-runtime-compatible-anthropic-e2e,
-          token-rotation-e2e, sandbox-survival-e2e,
-          openshell-gateway-upgrade-e2e,
-          issue-2478-crash-loop-recovery-e2e, hermes-e2e,
-          hermes-dashboard-e2e,
-          hermes-root-entrypoint-smoke-e2e,
-          openclaw-onboard-security-posture-e2e,
-          hermes-onboard-security-posture-e2e,
-          hermes-inference-switch-e2e, hermes-discord-e2e,
-          hermes-slack-e2e, sandbox-operations-e2e, inference-routing-e2e,
-          openclaw-inference-switch-e2e,
-          network-policy-e2e, state-backup-restore-e2e, tunnel-lifecycle-e2e, diagnostics-e2e,
-          credential-migration-e2e,
-          snapshot-commands-e2e, shields-config-e2e,
-          vm-driver-privileged-exec-routing-e2e, rebuild-openclaw-e2e,
-          upgrade-stale-sandbox-e2e, rebuild-hermes-e2e,
-          rebuild-hermes-stale-base-e2e, double-onboard-e2e,
-          onboard-repair-e2e, onboard-resume-e2e, onboard-negative-paths-e2e,
-          runtime-overrides-e2e,
-          credential-sanitization-e2e, telegram-injection-e2e,
-          overlayfs-autofix-e2e, device-auth-health-e2e,
-          launchable-smoke-e2e, gpu-e2e, gpu-double-onboard-e2e,
+          messaging-compatible-endpoint-e2e, sessions-agents-cli-e2e,
           channels-add-remove-e2e, channels-stop-start-e2e, brave-search-e2e,
-          sessions-agents-cli-e2e
+          kimi-inference-compat-e2e, bedrock-runtime-compatible-anthropic-e2e,
+          token-rotation-e2e, sandbox-survival-e2e, issue-2478-crash-loop-recovery-e2e,
+          hermes-e2e, hermes-dashboard-e2e, hermes-root-entrypoint-smoke-e2e,
+          openclaw-onboard-security-posture-e2e, hermes-onboard-security-posture-e2e,
+          hermes-inference-switch-e2e, hermes-discord-e2e, hermes-slack-e2e,
+          sandbox-operations-e2e, inference-routing-e2e, openclaw-inference-switch-e2e,
+          network-policy-e2e, state-backup-restore-e2e, tunnel-lifecycle-e2e,
+          diagnostics-e2e, credential-migration-e2e, snapshot-commands-e2e,
+          shields-config-e2e, vm-driver-privileged-exec-routing-e2e, rebuild-openclaw-e2e,
+          upgrade-stale-sandbox-e2e, openshell-gateway-upgrade-e2e, rebuild-hermes-e2e,
+          rebuild-hermes-stale-base-e2e, double-onboard-e2e, onboard-repair-e2e,
+          onboard-resume-e2e, onboard-negative-paths-e2e, runtime-overrides-e2e,
+          credential-sanitization-e2e, telegram-injection-e2e, overlayfs-autofix-e2e,
+          device-auth-health-e2e, launchable-smoke-e2e, gpu-e2e, gpu-double-onboard-e2e
         required: false
         type: string
         default: ""
@@ -221,6 +211,24 @@ jobs:
       env_json: '{"NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE":"1","NEMOCLAW_NON_INTERACTIVE":"1","NEMOCLAW_RECREATE_SANDBOX":"1","NEMOCLAW_SANDBOX_NAME":"e2e-cloud-inference"}'
       nvidia_api_key: true
     secrets: *nightly-e2e-default-secrets
+  agent-turn-latency-e2e:
+    if: >-
+      github.repository == 'NVIDIA/NemoClaw' && (github.event_name != 'workflow_dispatch' ||
+       inputs.jobs == '' ||
+       contains(format(',{0},', inputs.jobs), ',agent-turn-latency-e2e,'))
+    uses: ./.github/workflows/e2e-script.yaml
+    with:
+      ref: ${{ inputs.target_ref || github.ref }}
+      script: test/e2e/test-agent-turn-latency-e2e.sh
+      timeout_minutes: 120
+      artifact_name: "agent-turn-latency-logs"
+      artifact_path: |
+        /tmp/nemoclaw-e2e-openclaw-turn-latency-install.log
+        /tmp/nemoclaw-e2e-hermes-turn-latency-install.log
+        /tmp/nemoclaw-e2e-agent-turn-latency.json
+      env_json: '{"NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE":"1","NEMOCLAW_NON_INTERACTIVE":"1","NEMOCLAW_RECREATE_SANDBOX":"1","NEMOCLAW_TURN_LATENCY_MODEL":"nvidia/nemotron-3-ultra-550b-a55b"}'
+      nvidia_api_key: true
+    secrets: *nightly-e2e-default-secrets
   skill-agent-e2e:
     if: >-
       github.repository == 'NVIDIA/NemoClaw' && (github.event_name != 'workflow_dispatch' ||
@@ -2008,6 +2016,7 @@ jobs:
         cloud-e2e,
         cloud-onboard-e2e,
         cloud-inference-e2e,
+        agent-turn-latency-e2e,
         skill-agent-e2e,
         openclaw-skill-cli-e2e,
         docs-validation-e2e,
@@ -2118,6 +2127,7 @@ jobs:
         cloud-e2e,
         cloud-onboard-e2e,
         cloud-inference-e2e,
+        agent-turn-latency-e2e,
         skill-agent-e2e,
         openclaw-skill-cli-e2e,
         docs-validation-e2e,
@@ -2285,6 +2295,7 @@ jobs:
         cloud-e2e,
         cloud-onboard-e2e,
         cloud-inference-e2e,
+        agent-turn-latency-e2e,
         skill-agent-e2e,
         openclaw-skill-cli-e2e,
         docs-validation-e2e,

diff --git a/docs/inference/inference-options.mdx b/docs/inference/inference-options.mdx
@@ -64,7 +64,7 @@ The managed install/start vLLM entry appears by default on DGX Spark and DGX Sta
 
 | Option | Description | Curated models |
 |--------|-------------|----------------|
-| NVIDIA Endpoints | Routes to models hosted on [build.nvidia.com](https://build.nvidia.com). You can also enter any model ID from the catalog. Set `NVIDIA_API_KEY`. | Nemotron 3 Super 120B, GLM-5.1, MiniMax M2.7, GPT-OSS 120B, DeepSeek V4 Pro |
+| NVIDIA Endpoints | Routes to models hosted on [build.nvidia.com](https://build.nvidia.com). You can also enter any model ID from the catalog. Set `NVIDIA_API_KEY`. | Nemotron 3 Super 120B, Nemotron 3 Ultra 550B, GLM-5.1, MiniMax M2.7, GPT-OSS 120B, DeepSeek V4 Pro |
 | OpenAI | Routes to the OpenAI API. Set `OPENAI_API_KEY`. | `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-5.4-pro-2026-03-05` |
 | Other OpenAI-compatible endpoint | Routes to any server that implements `/v1/chat/completions`. NemoClaw uses `/v1/chat/completions` at runtime by default; set `NEMOCLAW_PREFERRED_API=openai-responses` to allow `/v1/responses` for proxies that implement it, such as some llama.cpp builds. The wizard prompts for a base URL and model name. Works with OpenRouter, LocalAI, llama.cpp, or any compatible proxy. When you enable Telegram messaging, onboarding also runs a bounded sandbox-side smoke check through `https://inference.local/v1/chat/completions`. Set `COMPATIBLE_API_KEY`. | You provide the model name. |
 | Anthropic | Routes to the Anthropic Messages API. Set `ANTHROPIC_API_KEY`. | `claude-sonnet-4-6`, `claude-haiku-4-5`, `claude-opus-4-6` |

diff --git a/src/lib/inference/config.test.ts b/src/lib/inference/config.test.ts
@@ -27,6 +27,7 @@ describe("inference selection config", () => {
   it("exposes the curated cloud model picker options", () => {
     expect(CLOUD_MODEL_OPTIONS.map((option: { id: string }) => option.id)).toEqual([
       "nvidia/nemotron-3-super-120b-a12b",
+      "nvidia/nemotron-3-ultra-550b-a55b",
       "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning",
       "z-ai/glm-5.1",
       "minimaxai/minimax-m2.7",

diff --git a/src/lib/inference/config.ts b/src/lib/inference/config.ts
@@ -47,6 +47,7 @@ export const HERMES_PROVIDER_MODEL_OPTIONS = [
 export const DEFAULT_HERMES_PROVIDER_MODEL = HERMES_PROVIDER_MODEL_OPTIONS[0];
 export const CLOUD_MODEL_OPTIONS = [
   { id: "nvidia/nemotron-3-super-120b-a12b", label: "Nemotron 3 Super 120B" },
+  { id: "nvidia/nemotron-3-ultra-550b-a55b", label: "Nemotron 3 Ultra 550B" },
   { id: "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning", label: "Nemotron 3 Nano Omni 30B" },
   { id: "z-ai/glm-5.1", label: "GLM-5" },
   { id: "minimaxai/minimax-m2.7", label: "MiniMax M2.7" },

diff --git a/src/lib/inference/model-prompts.test.ts b/src/lib/inference/model-prompts.test.ts
@@ -32,7 +32,7 @@ describe("model prompt helpers", () => {
   });
 
   it("returns DeepSeek V4 Pro from the default cloud model menu", async () => {
-    const promptFn = promptSequence(["7"]);
+    const promptFn = promptSequence(["8"]);
     const result = await promptCloudModel({
       promptFn,
       writeLine: vi.fn(),