fix(models): key Azure DeepSeek V4 Flash 0731 by its Foundry catalog id by yujonglee-berri · Pull Request #39341 · BerriAI/litellm

yujonglee-berri · 2026-09-02T15:12:02Z

TLDR

Problem this solves:

azure_ai/deepseek-v4-flash-0731 (added in feat: add Azure AI DeepSeek V4 Flash 0731 pricing #39023, priced in fix(models): registry audit 2026-09-01: openai realtime and long-context tiers, mistral aliases, voyage, xai, fireworks, together, scaleway, azure ai, govcloud, azure gov, cloudflare whisper, deprecation dates #39170) is keyed in lowercase, while Azure's catalog id, and the default name of a Foundry deployment, is DeepSeek-V4-Flash-0731
Cost lookups resolve either casing, but exact-key readers such as GET /model/deprecations and litellm.get_max_tokens miss deployments routed by the catalog id, so the 2026-12-03 retirement never surfaces for them

How it solves it:

Renames the entry to azure_ai/DeepSeek-V4-Flash-0731, the casing the rest of the azure_ai catalog uses (Phi, Llama, MAI, Cohere, FW-*)
Keeps every value from the base: Azure's "V4 Flash 0731 glbl" meters, the 1M/128K limits, the retirement date, and the capability flags

User Flow

Before: a developer routing azure_ai/DeepSeek-V4-Flash-0731 through the proxy is billed correctly but never sees the model's retirement date

The proxy admin adds a deployment with model: azure_ai/DeepSeek-V4-Flash-0731, their Azure AI Foundry endpoint and key, and restarts the proxy
The developer sends POST https://litellm-domain/v1/chat/completions with "model": "DeepSeek-V4-Flash-0731" and a one-line prompt, and gets 200 OK with 10 prompt tokens and 2 completion tokens
The response carries x-litellm-response-cost: 7.04e-06, Azure's 0731 rate, found through the case-insensitive registry lookup
GET https://litellm-domain/model/info reports the entry under key: azure_ai/deepseek-v4-flash-0731, an id that appears nowhere in the admin's config
GET https://litellm-domain/model/deprecations lists nothing, although Azure retires the model on 2026-12-03

After: the same requests cost the same and the retirement date shows up

The proxy admin adds a deployment with model: azure_ai/DeepSeek-V4-Flash-0731, their Azure AI Foundry endpoint and key, and restarts the proxy
The developer sends POST https://litellm-domain/v1/chat/completions with "model": "DeepSeek-V4-Flash-0731" and a one-line prompt, and gets 200 OK with 10 prompt tokens and 2 completion tokens
The response carries x-litellm-response-cost: 7.04e-06, unchanged
GET https://litellm-domain/model/info reports the entry under key: azure_ai/DeepSeek-V4-Flash-0731, the id in the admin's config
GET https://litellm-domain/model/deprecations lists the deployment under upcoming with deprecation_date: 2026-12-03

Relevant issues

Extracted from #38976 (closed). Follows #39023, which added the lowercase entry, and #39170, which priced it from Azure's 0731 meters; this PR only moves the entry to the catalog id

Linear ticket

Part of LIT-6555 (its Kimi-K3 half still needs its own PR)

Pre-Submission checklist

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

I have added meaningful tests (registry-only change; CI's model map schema check validates it, and the proof below is the live check)
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

Shared setup for both legs: a DeepSeek-V4-Flash-0731 deployment (GlobalStandard, version 2026-07-31) on an Azure AI Foundry resource in eastus, real Azure calls. Each leg is its own worktree and venv, booted with LITELLM_LOCAL_MODEL_COST_MAP=True python litellm/proxy/proxy_cli.py --config config.yaml --port <port> --num_workers 2, two uvicorn workers, no database, same config:

model_list:
  - model_name: DeepSeek-V4-Flash-0731
    litellm_params:
      model: azure_ai/DeepSeek-V4-Flash-0731
      api_base: https://<resource>.services.ai.azure.com
      api_key: os.environ/AZURE_AI_API_KEY
general_settings:
  master_key: os.environ/LITELLM_MASTER_KEY

Cost arithmetic on both legs: 10 prompt tokens x 4.4e-07 + 2 completion tokens x 1.32e-06 = 7.04e-06. The base already carries the 0731 meters through #39170, so this PR changes the key the values live under, not the values

Before (`4049a07`)

model info

Run (proxy on port 24024):

curl -s http://127.0.0.1:24024/model/info -H "Authorization: Bearer $LITELLM_MASTER_KEY" | jq '.data[0].model_info | {key, input_cost_per_token, output_cost_per_token, cache_read_input_token_cost, max_input_tokens, max_output_tokens, supports_function_calling, supports_tool_choice, supports_reasoning, supports_prompt_caching, deprecation_date}'

Observed:

{
  "key": "azure_ai/deepseek-v4-flash-0731",
  "input_cost_per_token": 4.4E-7,
  "output_cost_per_token": 0.00000132,
  "cache_read_input_token_cost": 1.4E-8,
  "max_input_tokens": 1000000,
  "max_output_tokens": 128000,
  "supports_function_calling": true,
  "supports_tool_choice": true,
  "supports_reasoning": true,
  "supports_prompt_caching": true,
  "deprecation_date": null
}

model deprecations

Run (proxy on port 24024):

curl -s http://127.0.0.1:24024/model/deprecations -H "Authorization: Bearer $LITELLM_MASTER_KEY" | jq -c '{deprecated, imminent, upcoming}'

Observed:

{"deprecated":[],"imminent":[],"upcoming":[]}

chat completions cost

Run (proxy on port 24024):

curl -s -D /dev/stderr http://127.0.0.1:24024/v1/chat/completions -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","messages":[{"role":"user","content":"Reply with exactly the word OK"}],"max_tokens":20}' 2> >(grep -i x-litellm-response-cost) | jq '{content: .choices[0].message.content, usage}'

Observed:

{
  "content": "OK",
  "usage": {
    "completion_tokens": 2,
    "prompt_tokens": 10,
    "total_tokens": 12,
    "audio_prompt_tokens": 0
  }
}
x-litellm-response-cost: 7.04e-06
x-litellm-response-cost-original: 7.04e-06
x-litellm-response-cost-discount-amount: 0.0
x-litellm-response-cost-margin-amount: 0.0
x-litellm-response-cost-margin-percent: 0.0
x-litellm-response-cost-input: 4.4e-06
x-litellm-response-cost-output: 2.64e-06
x-litellm-response-cost-tool-usage: 0.0

responses API cost

Run (proxy on port 24024):

curl -s -D /dev/stderr http://127.0.0.1:24024/v1/responses -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","input":"Reply with exactly the word OK","max_output_tokens":20}' 2> >(grep -i x-litellm-response-cost) | jq '{text: .output[-1].content[0].text, usage}'

Observed:

{
  "text": "OK",
  "usage": {
    "input_tokens": 10,
    "input_tokens_details": null,
    "output_tokens": 2,
    "output_tokens_details": null,
    "total_tokens": 12,
    "cost": null
  }
}
x-litellm-response-cost: 7.04e-06
x-litellm-response-cost-original: 7.04e-06
x-litellm-response-cost-discount-amount: 0.0
x-litellm-response-cost-margin-amount: 0.0
x-litellm-response-cost-margin-percent: 0.0
x-litellm-response-cost-input: 4.4e-06
x-litellm-response-cost-output: 2.64e-06
x-litellm-response-cost-tool-usage: 0.0

messages API cost

Run (proxy on port 24024):

curl -s -D /dev/stderr http://127.0.0.1:24024/v1/messages -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","max_tokens":20,"messages":[{"role":"user","content":"Reply with exactly the word OK"}]}' 2> >(grep -i x-litellm-response-cost) | jq '{text: .content[0].text, usage}'

Observed:

{
  "text": "OK",
  "usage": {
    "input_tokens": 10,
    "output_tokens": 2
  }
}
x-litellm-response-cost-original: 7.04e-06
x-litellm-response-cost-discount-amount: 0.0
x-litellm-response-cost-margin-amount: 0.0
x-litellm-response-cost-margin-percent: 0.0
x-litellm-response-cost-input: 4.4e-06
x-litellm-response-cost-output: 2.64e-06
x-litellm-response-cost-tool-usage: 0.0

tool calling

Run (proxy on port 24024):

curl -s http://127.0.0.1:24024/v1/chat/completions -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","messages":[{"role":"user","content":"What is the weather in Paris right now?"}],"tools":[{"type":"function","function":{"name":"get_weather","parameters":{"type":"object","properties":{"city":{"type":"string"}},"required":["city"]}}}],"tool_choice":"required"}' | jq '.choices[0].message.tool_calls'

Observed:

[
  {
    "index": 0,
    "function": {
      "arguments": "{\"city\": \"Paris\"}",
      "name": "get_weather"
    },
    "id": "call_8dcd8c56c6944c499c0a5992",
    "type": "function"
  }
]

After (`f38a1ec`)

model info

Run (proxy on port 32332):

curl -s http://127.0.0.1:32332/model/info -H "Authorization: Bearer $LITELLM_MASTER_KEY" | jq '.data[0].model_info | {key, input_cost_per_token, output_cost_per_token, cache_read_input_token_cost, max_input_tokens, max_output_tokens, supports_function_calling, supports_tool_choice, supports_reasoning, supports_prompt_caching, deprecation_date}'

Observed:

{
  "key": "azure_ai/DeepSeek-V4-Flash-0731",
  "input_cost_per_token": 4.4E-7,
  "output_cost_per_token": 0.00000132,
  "cache_read_input_token_cost": 1.4E-8,
  "max_input_tokens": 1000000,
  "max_output_tokens": 128000,
  "supports_function_calling": true,
  "supports_tool_choice": true,
  "supports_reasoning": true,
  "supports_prompt_caching": true,
  "deprecation_date": null
}

model deprecations

Run (proxy on port 32332):

curl -s http://127.0.0.1:32332/model/deprecations -H "Authorization: Bearer $LITELLM_MASTER_KEY" | jq -c '{deprecated, imminent, upcoming}'

Observed:

{"deprecated":[],"imminent":[],"upcoming":[{"model_name":"DeepSeek-V4-Flash-0731","litellm_model":"azure_ai/DeepSeek-V4-Flash-0731","deprecation_date":"2026-12-03","days_until_deprecation":92,"status":"upcoming","litellm_provider":"azure_ai"}]}

chat completions cost

Run (proxy on port 32332):

curl -s -D /dev/stderr http://127.0.0.1:32332/v1/chat/completions -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","messages":[{"role":"user","content":"Reply with exactly the word OK"}],"max_tokens":20}' 2> >(grep -i x-litellm-response-cost) | jq '{content: .choices[0].message.content, usage}'

Observed:

{
  "content": "OK",
  "usage": {
    "completion_tokens": 2,
    "prompt_tokens": 10,
    "total_tokens": 12,
    "audio_prompt_tokens": 0
  }
}
x-litellm-response-cost: 7.04e-06
x-litellm-response-cost-original: 7.04e-06
x-litellm-response-cost-discount-amount: 0.0
x-litellm-response-cost-margin-amount: 0.0
x-litellm-response-cost-margin-percent: 0.0
x-litellm-response-cost-input: 4.4e-06
x-litellm-response-cost-output: 2.64e-06
x-litellm-response-cost-tool-usage: 0.0

responses API cost

Run (proxy on port 32332):

curl -s -D /dev/stderr http://127.0.0.1:32332/v1/responses -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","input":"Reply with exactly the word OK","max_output_tokens":20}' 2> >(grep -i x-litellm-response-cost) | jq '{text: .output[-1].content[0].text, usage}'

Observed:

{
  "text": "OK",
  "usage": {
    "input_tokens": 10,
    "input_tokens_details": null,
    "output_tokens": 2,
    "output_tokens_details": null,
    "total_tokens": 12,
    "cost": null
  }
}
x-litellm-response-cost: 7.04e-06
x-litellm-response-cost-original: 7.04e-06
x-litellm-response-cost-discount-amount: 0.0
x-litellm-response-cost-margin-amount: 0.0
x-litellm-response-cost-margin-percent: 0.0
x-litellm-response-cost-input: 4.4e-06
x-litellm-response-cost-output: 2.64e-06
x-litellm-response-cost-tool-usage: 0.0

messages API cost

Run (proxy on port 32332):

curl -s -D /dev/stderr http://127.0.0.1:32332/v1/messages -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","max_tokens":20,"messages":[{"role":"user","content":"Reply with exactly the word OK"}]}' 2> >(grep -i x-litellm-response-cost) | jq '{text: .content[0].text, usage}'

Observed:

{
  "text": "OK",
  "usage": {
    "input_tokens": 10,
    "output_tokens": 2
  }
}
x-litellm-response-cost-original: 7.04e-06
x-litellm-response-cost-discount-amount: 0.0
x-litellm-response-cost-margin-amount: 0.0
x-litellm-response-cost-margin-percent: 0.0
x-litellm-response-cost-input: 4.4e-06
x-litellm-response-cost-output: 2.64e-06
x-litellm-response-cost-tool-usage: 0.0

tool calling

Run (proxy on port 32332):

curl -s http://127.0.0.1:32332/v1/chat/completions -H "Authorization: Bearer $LITELLM_MASTER_KEY" -H 'Content-Type: application/json' -d '{"model":"DeepSeek-V4-Flash-0731","messages":[{"role":"user","content":"What is the weather in Paris right now?"}],"tools":[{"type":"function","function":{"name":"get_weather","parameters":{"type":"object","properties":{"city":{"type":"string"}},"required":["city"]}}}],"tool_choice":"required"}' | jq '.choices[0].message.tool_calls'

Observed:

[
  {
    "index": 0,
    "function": {
      "arguments": "{\"city\": \"Paris\"}",
      "name": "get_weather"
    },
    "id": "call_6f61e3cc2db348de9f5d3c8f",
    "type": "function"
  }
]

Observed on both legs, not caused by this PR: /v1/messages omits the bare x-litellm-response-cost header and only sends the -original, -input, and -output variants

Type

🐛 Bug Fix

Caveats (if any)

Low

The lowercase key (shipped only in v1.101.0-dev.1) goes away
- get_model_info and cost tracking resolve either casing, so nothing billed changes for those callers
- litellm.get_max_tokens("azure_ai/deepseek-v4-flash-0731") is an exact-key lookup and now raises; the catalog id resolves instead
The other azure_ai DeepSeek keys (deepseek-r1, deepseek-v3-0324, deepseek-v3.1, deepseek-v4-flash) stay lowercase; moving them to catalog ids is out of scope here

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
f38a1ec passes /live-pr-risk

greptile-apps · 2026-09-02T15:13:51Z

Greptile Summary

The PR aligns the Azure DeepSeek V4 Flash 0731 registry key with its exact Foundry catalog identifier in both model-price registries.

Renames the primary registry key to azure_ai/DeepSeek-V4-Flash-0731.
Applies the identical key update to the packaged backup registry.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains.

No blocking failure remains.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Renames the Azure DeepSeek V4 Flash 0731 entry to the case-sensitive Foundry catalog identifier while retaining its pricing, limits, capabilities, and retirement metadata.
litellm/model_prices_and_context_window_backup.json	Mirrors the primary registry key change in the packaged backup model map.

_{Reviews (3): Last reviewed commit: "Merge branch 'litellm_internal_staging' ..." | Re-trigger Greptile}

codspeed-hq · 2026-09-02T15:14:36Z

Merging this PR will not alter performance

✅ 31 untouched benchmarks

_{Comparing litellm_azure_deepseek_v4_flash_0731 (f38a1ec) with litellm_internal_staging (b600f02)}

codecov · 2026-09-02T15:21:21Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

…nder the catalog id

mateo-berri · 2026-09-02T18:00:12Z

@greptileai

mateo-berri · 2026-09-02T18:04:42Z

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 a9d3a07. Configure here.}

…/litellm into litellm_azure_deepseek_v4_flash_0731 # Conflicts: # litellm/model_prices_and_context_window_backup.json # model_prices_and_context_window.json

mateo-berri · 2026-09-02T18:43:34Z

@greptileai

mateo-berri · 2026-09-02T18:44:02Z

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 f38a1ec. Configure here.}

mateo-berri

LGTM. Thanks!

feat(models): add Azure DeepSeek V4 Flash 0731

07cf9dc

yujonglee-berri requested a review from mateo-berri as a code owner September 2, 2026 15:12

yujonglee-berri mentioned this pull request Sep 2, 2026

feat: add Azure AI DeepSeek V4 Flash 0731 pricing #39023

Merged

fix(models): price Azure DeepSeek V4 Flash 0731 from its own meters u…

a9d3a07

…nder the catalog id

mateo-berri changed the title ~~feat(models): add Azure DeepSeek V4 Flash 0731~~ fix(models): price Azure DeepSeek V4 Flash 0731 from its own meters under the catalog id Sep 2, 2026

cursor Bot reviewed Sep 2, 2026

View reviewed changes

mateo-berri changed the title ~~fix(models): price Azure DeepSeek V4 Flash 0731 from its own meters under the catalog id~~ fix(models): key Azure DeepSeek V4 Flash 0731 by its Foundry catalog id Sep 2, 2026

Merge branch 'litellm_internal_staging' of https://github.com/BerriAI…

f38a1ec

…/litellm into litellm_azure_deepseek_v4_flash_0731 # Conflicts: # litellm/model_prices_and_context_window_backup.json # model_prices_and_context_window.json

cursor Bot reviewed Sep 2, 2026

View reviewed changes

mateo-berri approved these changes Sep 2, 2026

View reviewed changes

mateo-berri merged commit ba2e5d2 into litellm_internal_staging Sep 2, 2026
80 of 81 checks passed

mateo-berri deleted the litellm_azure_deepseek_v4_flash_0731 branch September 2, 2026 18:52

mateo-berri mentioned this pull request Sep 7, 2026

ci: report a shadow-mode PR risk tier from changed paths, diff size, and test delta #40171

Open

7 tasks

Navigation Menu

Uh oh!

Uh oh!

fix(models): key Azure DeepSeek V4 Flash 0731 by its Foundry catalog id - #39341

fix(models): key Azure DeepSeek V4 Flash 0731 by its Foundry catalog id#39341
mateo-berri merged 3 commits into
litellm_internal_stagingfrom
litellm_azure_deepseek_v4_flash_0731

yujonglee-berri commented Sep 2, 2026 •

edited by mateo-berri

Loading

Uh oh!

greptile-apps Bot commented Sep 2, 2026 •

edited

Loading

Important Files Changed

Uh oh!

codspeed-hq Bot commented Sep 2, 2026 •

edited

Loading

Uh oh!

codecov Bot commented Sep 2, 2026

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

cursor Bot left a comment

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

cursor Bot left a comment

Uh oh!

mateo-berri left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

yujonglee-berri commented Sep 2, 2026 • edited by mateo-berri Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

TLDR

User Flow

Relevant issues

Linear ticket

Pre-Submission checklist

Delays in PR merge?

Screenshots / Proof of Fix

Before (4049a07)

model info

model deprecations

chat completions cost

responses API cost

messages API cost

tool calling

After (f38a1ec)

model info

model deprecations

chat completions cost

responses API cost

messages API cost

tool calling

Type

Caveats (if any)

Low

Final Attestation

Uh oh!

greptile-apps Bot commented Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

codspeed-hq Bot commented Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Uh oh!

codecov Bot commented Sep 2, 2026

Codecov Report

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

mateo-berri commented Sep 2, 2026

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Uh oh!

mateo-berri left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

yujonglee-berri commented Sep 2, 2026 •

edited by mateo-berri

Loading

Before (`4049a07`)

After (`f38a1ec`)

greptile-apps Bot commented Sep 2, 2026 •

edited

Loading

codspeed-hq Bot commented Sep 2, 2026 •

edited

Loading