feat(cost-map): add azure/gpt-6-astra and azure/us/gpt-6-astra Foundry pricing

[mateo-berri]

TLDR

Problem this solves:

• Microsoft Foundry made gpt-6-astra GA on 2026-09-03
• LiteLLM has no azure/gpt-6-astra entry, so every Foundry call bills $0
• Spend, budgets, and model info all read zero for those deployments

How it solves it:

• Adds azure/gpt-6-astra at the Foundry Standard Global rates
• Adds azure/us/gpt-6-astra at the 10% US Data Zone uplift
• Long-context (272K+) and cache rates match the announcement
• Effort levels none and low through max, no minimal
• Foundry accepts reasoning_effort: none (unlike OpenAI), which unlocks temperature
• The Azure Responses config now reads the azure/ entry for that none gate instead of OpenAI's, so /v1/responses with reasoning.effort: none and a temperature stops 400ing
• Regression tests fail without the two entries and without the Responses fix

Sources: Foundry GA announcement (rates), the Foundry models page (context window, capabilities), and the Foundry API itself for the effort levels: a direct call with reasoning_effort: minimal answers Supported values are: 'none', 'low', 'medium', 'high', 'xhigh', 'max', and 'disabled', none returns 200 with zero reasoning tokens and accepts temperature: 0.2, while OpenAI's own API rejects none for gpt-6-astra (Supported values are: 'low', 'medium', 'high', and 'xhigh'). The default effort is not none (a call without reasoning_effort spends 67 reasoning tokens and rejects temperature: 0.2), so no default_reasoning_effort is declared

User Flow

Before: a proxy admin who deployed gpt-6-astra on Microsoft Foundry routes it through LiteLLM and every request bills $0, so spend tracking and budgets never see it

1. They add azure/gpt-6-astra to their config as model group gpt-6-astra and restart the proxy
2. They send GET https://litellm-domain/v1/model/info and the entry shows input_cost_per_token: 0, output_cost_per_token: 0, max_input_tokens: null, supports_reasoning: null
3. They send POST https://litellm-domain/v1/chat/completions with "model": "gpt-6-astra" and get a 200 with the answer, but the response headers say x-litellm-response-cost-input: 0.0 and x-litellm-response-cost-output: 0.0
4. The same happens on POST https://litellm-domain/v1/responses and POST https://litellm-domain/v1/messages
5. They send POST https://litellm-domain/spend/calculate with a 300K-token Astra response and get {"cost": 0.0}
6. https://litellm-domain/ui/?page=logs shows every Astra request at $0 and a team budget on that model never trips
7. They send POST https://litellm-domain/v1/chat/completions with "reasoning_effort": "none" and "temperature": 0.2, which Foundry itself accepts, and get a 400 Azure OpenAI does not support reasoning_effort='none' for this model
8. They send POST https://litellm-domain/v1/responses with "reasoning": {"effort": "none"} and "temperature": 0.2 and get a 400 gpt-6-astra doesn't support temperature=0.2 while reasoning is active; POST https://litellm-domain/v1/messages with "thinking": {"type": "disabled"} and the same temperature gets the step 7 400

After: the same deployment bills at the Foundry rates on every endpoint

1. They add azure/gpt-6-astra to their config as model group gpt-6-astra and restart the proxy
2. They send GET https://litellm-domain/v1/model/info and the entry shows input_cost_per_token: 0.00001, output_cost_per_token: 0.00005, max_input_tokens: 922000, supports_reasoning: true
3. They send POST https://litellm-domain/v1/chat/completions with "model": "gpt-6-astra" and get a 200 whose headers say x-litellm-response-cost: 0.00033 for 13 input and 4 output tokens
4. POST https://litellm-domain/v1/responses and POST https://litellm-domain/v1/messages carry the same non-zero cost headers
5. They send POST https://litellm-domain/spend/calculate with a 300K-token Astra response and get {"cost": 6.075}, the doubled long-context input rate plus 1.5x output
6. https://litellm-domain/ui/?page=logs shows real spend per Astra request and budgets on that model enforce
7. They send POST https://litellm-domain/v1/chat/completions with "reasoning_effort": "none" and "temperature": 0.2 and get a 200 with zero reasoning tokens and x-litellm-response-cost: 0.00033
8. They send POST https://litellm-domain/v1/responses with "reasoning": {"effort": "none"} and "temperature": 0.2 and get a 200 with zero reasoning tokens and x-litellm-response-cost: 0.00038; POST https://litellm-domain/v1/messages with "thinking": {"type": "disabled"} and the same temperature gets a 200 too

Relevant issues

Linear ticket

Resolves LIT-6997

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/test_litellm/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

Both legs boot the proxy from their own worktree and .venv with --num_workers 2 on a random free port, no database, reading the local cost map (LITELLM_LOCAL_MODEL_COST_MAP=True). The LLM calls go to a real gpt-6-astra GlobalStandard deployment on Microsoft Foundry. Same config, same commands, same order on both sides

Config:

model_list:
  - model_name: gpt-6-astra
    litellm_params:
      model: azure/gpt-6-astra
      api_key: os.environ/AZURE_DEVIN_FOUNDRY_QA_API_KEY
      api_base: os.environ/AZURE_DEVIN_FOUNDRY_QA_API_BASE
      api_version: "2025-04-01-preview"
  - model_name: gpt-6-astra-us
    litellm_params:
      model: azure/us/gpt-6-astra
      api_key: os.environ/AZURE_DEVIN_FOUNDRY_QA_API_KEY
      api_base: os.environ/AZURE_DEVIN_FOUNDRY_QA_API_BASE
      api_version: "2025-04-01-preview"

general_settings:
  master_key: sk-lit6997

Boot:

LITELLM_LOCAL_MODEL_COST_MAP=True python litellm/proxy/proxy_cli.py --config lit6997_config.yaml --port $PORT --num_workers 2 --use_v2_migration_resolver

Before (d23bec8)

/v1/model/info

1. curl -s http://localhost:20710/v1/model/info -H 'Authorization: Bearer sk-lit6997' | jq -c '.data[] | {model_name, litellm_model: .litellm_params.model, input_cost_per_token: .model_info.input_cost_per_token, input_above_272k: .model_info.input_cost_per_token_above_272k_tokens, output_cost_per_token: .model_info.output_cost_per_token, cache_read: .model_info.cache_read_input_token_cost, cache_creation: .model_info.cache_creation_input_token_cost, max_input_tokens: .model_info.max_input_tokens, supports_reasoning: .model_info.supports_reasoning}'
2. Output:

{"model_name":"gpt-6-astra","litellm_model":"azure/gpt-6-astra","input_cost_per_token":0,"input_above_272k":null,"output_cost_per_token":0,"cache_read":0.000001,"cache_creation":0.0000125,"max_input_tokens":null,"supports_reasoning":null}
{"model_name":"gpt-6-astra-us","litellm_model":"azure/us/gpt-6-astra","input_cost_per_token":0,"input_above_272k":null,"output_cost_per_token":0,"cache_read":null,"cache_creation":null,"max_input_tokens":null,"supports_reasoning":null}

/model_group/info

1. curl -s http://localhost:20710/model_group/info -H 'Authorization: Bearer sk-lit6997' | jq -c '.data[] | {model_group, supports_reasoning, supported_reasoning_efforts, max_input_tokens, input_cost_per_token, output_cost_per_token}'
2. Output:

{"model_group":"gpt-6-astra","supports_reasoning":false,"supported_reasoning_efforts":["low","medium","high","xhigh","max"],"max_input_tokens":null,"input_cost_per_token":0.0,"output_cost_per_token":0.0}
{"model_group":"gpt-6-astra-us","supports_reasoning":false,"supported_reasoning_efforts":null,"max_input_tokens":null,"input_cost_per_token":0.0,"output_cost_per_token":0.0}

/spend/calculate

1. Standard request, 1000 prompt tokens of which 200 cached, 100 completion tokens: curl -s -X POST http://localhost:20710/spend/calculate -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"completion_response":{"id":"chatcmpl-azure-astra-1","object":"chat.completion","created":1788500000,"model":"azure/gpt-6-astra","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1000,"completion_tokens":100,"total_tokens":1100,"prompt_tokens_details":{"cached_tokens":200}}}}'
2. Output: {"cost":0.00019999999999999998} (only the 200 cached tokens bill, through a generic cache fallback; the 800 text tokens and 100 output tokens are free)
3. Long context, 300000 prompt tokens, 1000 completion tokens: curl -s -X POST http://localhost:20710/spend/calculate -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"completion_response":{"id":"chatcmpl-azure-astra-2","object":"chat.completion","created":1788500000,"model":"azure/gpt-6-astra","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}],"usage":{"prompt_tokens":300000,"completion_tokens":1000,"total_tokens":301000}}}'
4. Output: {"cost":0.0}
5. US Data Zone, 1000 prompt tokens, 100 completion tokens: curl -s -X POST http://localhost:20710/spend/calculate -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"completion_response":{"id":"chatcmpl-azure-astra-3","object":"chat.completion","created":1788500000,"model":"azure/us/gpt-6-astra","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1000,"completion_tokens":100,"total_tokens":1100}}}'
6. Output: {"cost":0.0}

POST /v1/chat/completions (real Foundry call)

1. curl -s -D - http://localhost:20710/v1/chat/completions -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","messages":[{"role":"user","content":"Reply with the single word astra"}],"reasoning_effort":"low"}'
2. Output (headers filtered to status, upstream, and cost; body reduced with jq):

HTTP/1.1 200 OK
x-litellm-model-api-base: https://devinfoundryqa33856.services.ai.azure.com
x-litellm-response-cost-input: 0.0
x-litellm-response-cost-output: 0.0
{"model":"gpt-6-astra","content":"astra","usage":{"prompt_tokens":13,"completion_tokens":4}}

POST /v1/responses (real Foundry call)

1. curl -s -D - http://localhost:20710/v1/responses -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","input":"Reply with the single word astra","reasoning":{"effort":"low"}}'
2. Output:

HTTP/1.1 200 OK
x-litellm-model-api-base: https://devinfoundryqa33856.services.ai.azure.com
x-litellm-response-cost-input: 0.0
x-litellm-response-cost-output: 0.0
{"model":"gpt-6-astra","text":"astra","usage":{"input_tokens":13,"output_tokens":5}}

POST /v1/messages (real Foundry call)

1. curl -s -D - http://localhost:20710/v1/messages -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","max_tokens":256,"messages":[{"role":"user","content":"Reply with the single word astra"}]}'
2. Output:

HTTP/1.1 200 OK
x-litellm-response-cost-input: 0.0
x-litellm-response-cost-output: 0.0
{"model":"gpt-6-astra","text":"astra","usage":{"input_tokens":13,"output_tokens":4}}

POST /v1/chat/completions with reasoning_effort none and temperature 0.2 (real Foundry call)

1. curl -s -D - http://localhost:20710/v1/chat/completions -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","messages":[{"role":"user","content":"Reply with the single word astra"}],"reasoning_effort":"none","temperature":0.2}'
2. Output (LiteLLM rejects it before Foundry sees it; the same request sent straight to Foundry returns 200):

HTTP/1.1 400 Bad Request
x-litellm-response-cost: 0
{"error":"litellm.UnsupportedParamsError: Azure OpenAI does not support reasoning_effort='none' for this model. Supported values are: 'low', 'medium', and 'high'. To drop"}

3. Same request with "model":"gpt-6-astra-us": HTTP/1.1 400 Bad Request with the same error

POST /v1/responses with reasoning.effort none and temperature 0.2 (real Foundry call)

1. curl -s -D - http://localhost:20710/v1/responses -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","input":"Reply with the single word astra","reasoning":{"effort":"none"},"temperature":0.2}'
2. Output (LiteLLM rejects it before Foundry sees it):

HTTP/1.1 400 Bad Request
x-litellm-response-cost: 0
{"error":"litellm.UnsupportedParamsError: gpt-6-astra doesn't support temperature=0.2 while reasoning is active. Only temperature=1 is supported unless reasoning.effort resolves to 'none', either set explicitly"}

3. curl -s -D - http://localhost:20710/v1/messages -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","max_tokens":64,"messages":[{"role":"user","content":"Reply with the single word astra"}],"thinking":{"type":"disabled"},"temperature":0.2}'
4. Output:

HTTP/1.1 400 Bad Request
x-litellm-response-cost: 0
{"error":"litellm.UnsupportedParamsError: Azure OpenAI does not support reasoning_effort='none' for this model. Supported values are: 'low', 'medium', and 'high'. To drop this parameter, set `litellm.drop_param"}

After (c02f198)

/v1/model/info

1. curl -s http://localhost:27368/v1/model/info -H 'Authorization: Bearer sk-lit6997' | jq -c '.data[] | {model_name, litellm_model: .litellm_params.model, input_cost_per_token: .model_info.input_cost_per_token, input_above_272k: .model_info.input_cost_per_token_above_272k_tokens, output_cost_per_token: .model_info.output_cost_per_token, cache_read: .model_info.cache_read_input_token_cost, cache_creation: .model_info.cache_creation_input_token_cost, max_input_tokens: .model_info.max_input_tokens, supports_reasoning: .model_info.supports_reasoning}'
2. Output:

{"model_name":"gpt-6-astra","litellm_model":"azure/gpt-6-astra","input_cost_per_token":0.00001,"input_above_272k":0.00002,"output_cost_per_token":0.00005,"cache_read":0.000001,"cache_creation":0.0000125,"max_input_tokens":922000,"supports_reasoning":true}
{"model_name":"gpt-6-astra-us","litellm_model":"azure/us/gpt-6-astra","input_cost_per_token":0.000011,"input_above_272k":0.000022,"output_cost_per_token":0.000055,"cache_read":0.0000011,"cache_creation":0.00001375,"max_input_tokens":922000,"supports_reasoning":true}

/model_group/info

1. curl -s http://localhost:27368/model_group/info -H 'Authorization: Bearer sk-lit6997' | jq -c '.data[] | {model_group, supports_reasoning, supported_reasoning_efforts, max_input_tokens, input_cost_per_token, output_cost_per_token}'
2. Output:

{"model_group":"gpt-6-astra","supports_reasoning":true,"supported_reasoning_efforts":["none","low","medium","high","xhigh","max"],"max_input_tokens":922000.0,"input_cost_per_token":0.00001,"output_cost_per_token":0.00005}
{"model_group":"gpt-6-astra-us","supports_reasoning":true,"supported_reasoning_efforts":["none","low","medium","high","xhigh","max"],"max_input_tokens":922000.0,"input_cost_per_token":0.000011,"output_cost_per_token":0.000055}

/spend/calculate

1. Standard request, 1000 prompt tokens of which 200 cached, 100 completion tokens: curl -s -X POST http://localhost:27368/spend/calculate -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"completion_response":{"id":"chatcmpl-azure-astra-1","object":"chat.completion","created":1788500000,"model":"azure/gpt-6-astra","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1000,"completion_tokens":100,"total_tokens":1100,"prompt_tokens_details":{"cached_tokens":200}}}}'
2. Output: {"cost":0.0132} = 800 x $10/M + 200 x $1/M + 100 x $50/M
3. Long context, 300000 prompt tokens, 1000 completion tokens: curl -s -X POST http://localhost:27368/spend/calculate -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"completion_response":{"id":"chatcmpl-azure-astra-2","object":"chat.completion","created":1788500000,"model":"azure/gpt-6-astra","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}],"usage":{"prompt_tokens":300000,"completion_tokens":1000,"total_tokens":301000}}}'
4. Output: {"cost":6.075000000000001} = 300000 x $20/M + 1000 x $75/M
5. US Data Zone, 1000 prompt tokens, 100 completion tokens: curl -s -X POST http://localhost:27368/spend/calculate -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"completion_response":{"id":"chatcmpl-azure-astra-3","object":"chat.completion","created":1788500000,"model":"azure/us/gpt-6-astra","choices":[{"index":0,"message":{"role":"assistant","content":"hi"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1000,"completion_tokens":100,"total_tokens":1100}}}'
6. Output: {"cost":0.0165} = 1000 x $11/M + 100 x $55/M

POST /v1/chat/completions (real Foundry call)

1. curl -s -D - http://localhost:27368/v1/chat/completions -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","messages":[{"role":"user","content":"Reply with the single word astra"}],"reasoning_effort":"low"}'
2. Output, $0.00033 = 13 x $10/M + 4 x $50/M:

HTTP/1.1 200 OK
x-litellm-model-api-base: https://devinfoundryqa33856.services.ai.azure.com
x-litellm-response-cost: 0.00033
x-litellm-response-cost-input: 0.00013000000000000002
x-litellm-response-cost-output: 0.0002
{"model":"gpt-6-astra","content":"astra","usage":{"prompt_tokens":13,"completion_tokens":4}}

POST /v1/responses (real Foundry call)

1. curl -s -D - http://localhost:27368/v1/responses -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","input":"Reply with the single word astra","reasoning":{"effort":"low"}}'
2. Output, $0.00038 = 13 x $10/M + 5 x $50/M:

HTTP/1.1 200 OK
x-litellm-model-api-base: https://devinfoundryqa33856.services.ai.azure.com
x-litellm-response-cost: 0.00038
x-litellm-response-cost-input: 0.00013000000000000002
x-litellm-response-cost-output: 0.00025
{"model":"gpt-6-astra","text":"astra","usage":{"input_tokens":13,"output_tokens":5}}

POST /v1/messages (real Foundry call)

1. curl -s -D - http://localhost:27368/v1/messages -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","max_tokens":256,"messages":[{"role":"user","content":"Reply with the single word astra"}]}'
2. Output, $0.00033 = 13 x $10/M + 4 x $50/M:

HTTP/1.1 200 OK
x-litellm-response-cost: 0.00033
x-litellm-response-cost-input: 0.00013000000000000002
x-litellm-response-cost-output: 0.0002
{"model":"gpt-6-astra","text":"astra","usage":{"input_tokens":13,"output_tokens":4}}

POST /v1/chat/completions with reasoning_effort none and temperature 0.2 (real Foundry call)

1. curl -s -D - http://localhost:27368/v1/chat/completions -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","messages":[{"role":"user","content":"Reply with the single word astra"}],"reasoning_effort":"none","temperature":0.2}'
2. Output:

HTTP/1.1 200 OK
x-litellm-model-api-base: https://devinfoundryqa33856.services.ai.azure.com
x-litellm-response-cost: 0.00033
x-litellm-response-cost-input: 0.00013000000000000002
x-litellm-response-cost-output: 0.0002
{"model":"gpt-6-astra","content":"astra","usage":{"prompt_tokens":13,"completion_tokens":4,"reasoning_tokens":0}}

3. Same request with "model":"gpt-6-astra-us" (the 404 is Foundry's: azure/us/gpt-6-astra is a cost-tracking entry, not a deployment name, see the caveats):

HTTP/1.1 404 Not Found
x-litellm-response-cost: 0
{"error":"litellm.NotFoundError: AzureException NotFoundError - Resource not found. Received Model Group=gpt-6-astra-us\nAvailable Model Group Fallbacks=None"}

POST /v1/responses with reasoning.effort none and temperature 0.2 (real Foundry call)

1. curl -s -D - http://localhost:27368/v1/responses -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","input":"Reply with the single word astra","reasoning":{"effort":"none"},"temperature":0.2}'
2. Output:

HTTP/1.1 200 OK
x-litellm-response-cost: 0.00038
{"model":"gpt-6-astra","status":"completed","text":"astra","reasoning_tokens":0,"temperature":0.2}

3. curl -s -D - http://localhost:27368/v1/messages -H 'Authorization: Bearer sk-lit6997' -H 'Content-Type: application/json' -d '{"model":"gpt-6-astra","max_tokens":64,"messages":[{"role":"user","content":"Reply with the single word astra"}],"thinking":{"type":"disabled"},"temperature":0.2}'
4. Output:

HTTP/1.1 200 OK
x-litellm-response-cost: 0.00033
{"model":"gpt-6-astra","text":"astra","usage":{"input_tokens":13,"output_tokens":4}}

Type

🆕 New Feature

Caveats (if any)

Low

• No _priority rates: Microsoft has not published a fast-mode price for Astra on Foundry
  • a service_tier: priority request bills at the standard rate until they do
  • guessing 2x from the gpt-5.6 precedent would be a made-up number, so left out
• No azure/eu/gpt-6-astra and no batch rates: Foundry sells Astra as Standard Global and Standard US Data Zone only, with no batch SKU
• The Azure retail prices API returns no Astra meters yet, so the rates come from the GA announcement
• reasoning_effort: disabled, which Foundry also lists, has no cost-map flag in LiteLLM and passes through untouched
• azure/us/gpt-6-astra is a cost-tracking entry (model_info.base_model), not a deployment name: a litellm_params.model of azure/us/gpt-6-astra hits Foundry with the deployment us/gpt-6-astra and 404s, same as every other azure/us/ entry
• The Responses gate change was swept over all 43 azure/ gpt-5-series deployment names in the cost map: the azure/ lookup already falls back to the bare OpenAI entry when the Azure entry says nothing ([Bug]: supports_response_schema returns incorrect metadata for DeepSeek models due to unreachable complete configuration #20885), so the only other deployments whose gate moves are gpt-5.1-chat and gpt-5.1-chat-2025-11-13, and they move the right way
  • the OpenAI map has no bare gpt-5.1-chat entry (it is gpt-5.1-chat-latest), so Azure Responses used to refuse or drop temperature for those deployments while the Azure chat surface already forwarded it; now both surfaces read the same azure/gpt-5.1-chat entry (none supported, none default) and keep it
• CI reds at the tip, none PR-caused, all optional under the base ruleset:
  • ci/circleci: proxy_store_model_in_db_tests fails test_chat_completion_bad_model_with_spend_logs (spend log model_group comes back empty on a nonexistent model), red on every staging scheduled pipeline since 89084 and tracked in LIT-6949; nothing in that test reads the cost map
  • proxy-infra / Run tests (Python 3.10) is the known staging red tracked in LIT-6947
  • Vertex AI / Run tests (Python 3.13) died before running a test (Failed to resolve action download info, a GitHub Actions download timeout); green on the rerun from failed

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR
• c02f198 passes /live-pr-risk

---

Note

Low Risk
Mostly cost-map and parameter-validation wiring; the Responses override affects Azure GPT-5-series temperature gates broadly but is aligned with existing Azure chat behavior.

Overview
Adds azure/gpt-6-astra and azure/us/gpt-6-astra to the bundled cost maps so Microsoft Foundry deployments bill at the published rates (global vs 10% US Data Zone), including long-context (272K+) and cache token pricing, plus capability flags for reasoning (none through max, not minimal).

AzureOpenAIResponsesAPIConfig now delegates _supports_reasoning_effort_none and _effort_resolves_to_none to AzureOpenAIGPT5Config, so /v1/responses temperature gating follows the azure/ cost-map entry for bare deployment names like gpt-6-astra instead of OpenAI’s map—allowing reasoning.effort: "none" with non-default temperature on Foundry where OpenAI rejects none.

Regression tests cover Foundry price-sheet math, chat/responses param mapping, and advertised reasoning efforts for both model keys.

^{Reviewed by Cursor Bugbot for commit c02f198. Bugbot is set up for automated code reviews on this repo. Configure here.}

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing litellm_azure_gpt_6_astra (c02f198) with litellm_internal_staging (59d42d3)}

[greptile-apps]

Greptile Summary

Adds Microsoft Foundry pricing and capability metadata for global and US Data Zone GPT-6 Astra deployments.

• Adds standard, cached, and long-context token rates to both bundled cost maps.
• Makes Azure Responses reasoning and temperature validation use Azure-specific model capabilities.
• Adds regression coverage for cost calculation, supported reasoning efforts, and request parameter mapping.

Confidence Score: 5/5

The PR appears safe to merge, with no outstanding actionable findings.

The cost-map entries are synchronized and covered for standard, cached, US Data Zone, and long-context billing, while the Azure Responses change reuses existing Azure capability lookup behavior. The prior reservation finding was manually resolved after Greptile agreed that it was a pre-existing platform-wide limitation rather than a regression from this PR.

Important Files Changed

Filename	Overview
litellm/llms/azure/responses/transformation.py	Delegates reasoning-effort capability checks to the existing Azure GPT-5-family configuration.
model_prices_and_context_window.json	Adds global and US Data Zone GPT-6 Astra pricing, context limits, endpoints, modalities, and capability flags.
litellm/model_prices_and_context_window_backup.json	Keeps the backup cost map synchronized with the primary GPT-6 Astra entries.
tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py	Verifies standard and long-context cost calculations for both Foundry pricing zones.
tests/test_litellm/llms/azure/chat/test_azure_gpt5_transformation.py	Verifies Azure Astra accepts none reasoning with temperature and rejects minimal effort.
tests/test_litellm/llms/azure/response/test_azure_transformation.py	Verifies Azure Responses temperature gating for none and active reasoning.
tests/test_litellm/router_utils/test_reasoning_effort_capability.py	Verifies both Foundry model entries advertise the intended reasoning-effort levels.

_{Reviews (3): Last reviewed commit: "fix(azure): responses none-effort temper..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[mateo-berri]

@greptileai

[mateo-berri]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit c02f198. Configure here.}

[mateo-berri]

@greptileai