Complete Claude 4.6 support alignment with compatibility aliases by PeterDaveHello · Pull Request #21413 · BerriAI/litellm

PeterDaveHello · 2026-02-17T21:37:35Z

Pre-Submission checklist

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

I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
My PR passes all unit tests on make test-unit
- make test-unit run hit a local litellm_enterprise editable-install mismatch (not related to this PR's changes). Focused validation tests passed: 123 passed.
My PR's scope is as isolated as possible, it only solves 1 specific problem
I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

CI (LiteLLM team)

CI status guideline:

50-55 passing tests: main is stable with minor issues.

45-49 passing tests: acceptable but needs attention

<= 40 passing tests: unstable; be careful with your merges and assess the risk.

Branch creation CI run
Link:
CI run for the last commit
Link:
Merge / cherry-pick CI run
Links:

Type

🐛 Bug Fix
✅ Test

Changes

Align Claude 4.6 model handling in Anthropic transformation logic:
- recognize Sonnet 4.6 aliases (sonnet-4-6, sonnet_4_6, sonnet-4.6)
- recognize dotted Opus 4.6 aliases (opus-4.6)
- map reasoning_effort to adaptive thinking for Sonnet 4.6 (matching Opus 4.6 behavior)
- keep output_config.effort=max restricted to Opus 4.6
Align Bedrock Claude 4.6 regional model entries with canonical IDs while preserving compatibility aliases:
- Opus 4.6: include jp and apac, and keep au alias for compatibility
- Sonnet 4.6: include canonical jp and keep apac alias for compatibility
Keep model price maps in sync:
- model_prices_and_context_window.json
- litellm/model_prices_and_context_window_backup.json
Extend and minimally adjust focused tests:
- tests/litellm/llms/anthropic/test_anthropic_reasoning_effort.py
- tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py
- tests/test_litellm/test_claude_opus_4_6_config.py

Validation run

Executed:

poetry run pytest tests/litellm/llms/anthropic/test_anthropic_reasoning_effort.py tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py tests/test_litellm/test_claude_opus_4_6_config.py

Result:

123 passed

vercel · 2026-02-17T21:37:40Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 18, 2026 0:59am

PeterDaveHello · 2026-02-17T21:38:01Z

@greptileai

greptile-apps · 2026-02-17T21:42:04Z

Greptile Summary

This PR extends Claude 4.6 support by adding Sonnet 4.6 model detection and wiring it into the existing Anthropic transformation logic. Key changes:

Adds _is_claude_sonnet_4_6() helper for detecting Sonnet 4.6 model aliases (sonnet-4-6, sonnet_4_6, sonnet-4.6)
Adds dotted alias opus-4.6 to the existing Opus 4.6 detector
Maps reasoning_effort to adaptive thinking for Sonnet 4.6, matching Opus 4.6 behavior
Adds Sonnet 4.6 aliases to the native structured output (output_format) model list
Correctly keeps effort='max' restricted to Opus 4.6 only
Adds Bedrock regional entries (jp., apac.) for Opus 4.6 and Sonnet 4.6 in both pricing JSON files
Extends tests thoroughly covering adaptive thinking, effort validation, and structured output for all new aliases

Confidence Score: 4/5

This PR is safe to merge — it adds model aliases and routing logic with good test coverage and no behavioral regressions.
The changes are well-scoped, follow existing patterns in the codebase, and include thorough test coverage (123 tests passing per the author). The transformation logic correctly restricts effort='max' to Opus 4.6 only while extending adaptive thinking to Sonnet 4.6. One minor style inconsistency noted in the response_format model list (underscore aliases only added for 4.6, not retroactively for 4.5), but this is pre-existing and low-risk.
litellm/llms/anthropic/chat/transformation.py — core logic changes for model detection and routing.

Important Files Changed

Filename	Overview
litellm/llms/anthropic/chat/transformation.py	Adds `_is_claude_sonnet_4_6` helper and wires Sonnet 4.6 into adaptive thinking and native structured output paths. Correctly restricts `effort='max'` to Opus 4.6 only. Adds underscore aliases for 4.6 models in response_format model set.
model_prices_and_context_window.json	Adds `jp.` and `apac.` regional entries for Opus 4.6 and `jp.` for Sonnet 4.6. Pricing and capabilities match existing regional entries for these models.
litellm/model_prices_and_context_window_backup.json	Backup file kept in sync with main model prices JSON. Same regional entries added.
tests/litellm/llms/anthropic/test_anthropic_reasoning_effort.py	Adds tests for Opus 4.6 dotted alias and Sonnet 4.6 adaptive thinking. All mock-only tests, no network calls.
tests/test_litellm/llms/anthropic/chat/test_anthropic_chat_transformation.py	Extends structured output and effort validation tests to cover Sonnet 4.6. Adds test for reasoning_effort adaptive mapping for Sonnet 4.6. All tests are properly mocked.
tests/test_litellm/test_claude_opus_4_6_config.py	Updates region validation test to expect `jp.` and `apac.` entries in addition to existing `au.` alias. Adds Sonnet 4.6 regional registration assertions.

Flowchart

flowchart TD
    A[incoming model string] --> B{_is_claude_opus_4_6?}
    A --> C{_is_claude_sonnet_4_6?}
    
    B -->|Yes| D[adaptive thinking]
    C -->|Yes| D
    B -->|No| E{other reasoning model?}
    C -->|No| E
    E -->|Yes| F[budget-based thinking]
    E -->|No| G[no thinking param]

    A --> H{response_format check}
    H -->|model matches 4.5/4.6 set| I[native output_format]
    H -->|other model| J[tool-based workaround]

    A --> K{output_config.effort == max?}
    K -->|Yes| B2{_is_claude_opus_4_6?}
    B2 -->|Yes| L[allow max effort]
    B2 -->|No| M[raise ValueError]
    K -->|No| N[allow effort value]

_{Last reviewed commit: 0bfd2b0}

greptile-apps

_{6 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0bfd2b0b44

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-02-17T21:44:24Z

litellm/llms/anthropic/chat/transformation.py

+            AnthropicConfig._is_claude_opus_4_6(model)
+            or AnthropicConfig._is_claude_sonnet_4_6(model)


Allow reasoning_effort for dotted Claude 4.6 aliases

This adds dotted alias handling to _map_reasoning_effort, but the normal completion flow validates params against get_supported_openai_params() before this mapping runs, and that list only includes reasoning_effort when supports_reasoning(model) is true. Because the Anthropic model map does not include claude-opus-4.6* / claude-sonnet-4.6* keys, those aliases still fail support detection and reasoning_effort is rejected as unsupported in utils._check_valid_arg, so the new dotted-alias reasoning mapping is unreachable in production requests.

Useful? React with 👍 / 👎.

Align Claude Sonnet 4.6 and Opus 4.6 support with official Anthropic and Bedrock model behavior while preserving existing caller compatibility. Update Anthropic transformation logic to recognize Sonnet 4.6 and dotted Opus 4.6 aliases in reasoning and structured output handling. Keep effort=max restricted to Opus 4.6 only. Align Bedrock regional mappings to include canonical regional IDs and keep legacy aliases to avoid breaking existing integrations: - Opus 4.6: keep jp/apac canonical coverage and preserve au alias - Sonnet 4.6: keep jp canonical and preserve apac alias Keep test changes minimal by extending existing Claude 4.6 config tests with only required assertions for new aliases and regional parity.

PeterDaveHello · 2026-02-18T18:13:25Z

Close due to #21448, opened a new #21487 to replace this one.

vercel bot deployed to Preview February 17, 2026 21:39 View deployment

greptile-apps bot reviewed Feb 17, 2026

View reviewed changes

chatgpt-codex-connector bot reviewed Feb 17, 2026

View reviewed changes

PeterDaveHello force-pushed the fix/complete-sonnet-opus-4-6-support branch from 0bfd2b0 to f45ed3e Compare February 17, 2026 22:14

vercel bot deployed to Preview February 17, 2026 22:15 View deployment

PeterDaveHello force-pushed the fix/complete-sonnet-opus-4-6-support branch from f45ed3e to c088c58 Compare February 18, 2026 12:57

vercel bot deployed to Preview February 18, 2026 12:59 View deployment

PeterDaveHello closed this Feb 18, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Complete Claude 4.6 support alignment with compatibility aliases#21413

Complete Claude 4.6 support alignment with compatibility aliases#21413
PeterDaveHello wants to merge 1 commit intoBerriAI:mainfrom
PeterDaveHelloKitchen:fix/complete-sonnet-opus-4-6-support

PeterDaveHello commented Feb 17, 2026

Uh oh!

vercel bot commented Feb 17, 2026 •

edited

Loading

Uh oh!

PeterDaveHello commented Feb 17, 2026

Uh oh!

greptile-apps bot commented Feb 17, 2026

Important Files Changed

Uh oh!

greptile-apps bot left a comment

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Feb 17, 2026

Uh oh!

PeterDaveHello commented Feb 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		AnthropicConfig._is_claude_opus_4_6(model)
		or AnthropicConfig._is_claude_sonnet_4_6(model)

Uh oh!

Conversation

PeterDaveHello commented Feb 17, 2026

Pre-Submission checklist

CI (LiteLLM team)

Type

Changes

Validation run

Uh oh!

vercel bot commented Feb 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

PeterDaveHello commented Feb 17, 2026

Uh oh!

greptile-apps bot commented Feb 17, 2026

Greptile Summary

Confidence Score: 4/5

Important Files Changed

Flowchart

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Feb 17, 2026

Choose a reason for hiding this comment

Uh oh!

PeterDaveHello commented Feb 18, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

vercel bot commented Feb 17, 2026 •

edited

Loading