Litellm day 0 opus 4.7 support

[Sameerlite]

Relevant issues

Pre-Submission checklist

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

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• 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

Delays in PR merge?

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

Screenshots / Proof of Fix

Type

🆕 New Feature

Changes

• Added support for Opus 4.7 across all providers
• Tool search is not supported from bedrock

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Apr 16, 2026 4:29pm

[greptile-apps]

Greptile Summary

This PR adds day-0 support for Claude Opus 4.7 across Anthropic, Bedrock (Converse + Invoke/Messages), Vertex AI, and Azure AI providers, including new pricing/capability entries in the model map, adaptive thinking handling, and correctly excluding tool-search on Bedrock Invoke for this model.

• effort='max' validation is hardcoded in _apply_output_config (self._is_opus_4_6_model(model) instead of self._supports_effort_level(model, \"max\")), directly contradicting the data-driven xhigh check right below it. The JSON schema already declares supports_max_reasoning_effort as a valid boolean field — adding it to the Opus 4.6 entries and using _supports_effort_level(model, \"max\") would make future model support a pure data change, consistent with the project's own rule.

Confidence Score: 4/5

Safe to merge after addressing the max effort validation inconsistency — no correctness bugs in the Opus 4.7 routing itself.

One P1 finding: effort='max' uses a hardcoded _is_opus_4_6_model() check while effort='xhigh' directly below uses the data-driven _supports_effort_level() pattern. The JSON schema already declares supports_max_reasoning_effort as a valid field, making the fix straightforward. All other changes (pricing entries, adaptive thinking routing, Bedrock tool-search exclusion, tests) look correct.

litellm/llms/anthropic/chat/transformation.py — _apply_output_config max effort guard

Important Files Changed

Filename	Overview
litellm/llms/anthropic/chat/transformation.py	Adds _is_opus_4_7_model() and _is_claude_4_7_model() helpers and wires them into reasoning-effort mapping and output_config routing; effort='max' guard remains hardcoded against Opus 4.6 instead of using the data-driven _supports_effort_level pattern already used for xhigh.
litellm/llms/anthropic/common_utils.py	Adds _is_claude_4_7_model() and _is_adaptive_thinking_model() string-match helpers; wires Opus 4.7 into beta-header logic. Hardcoded pattern is consistent with the existing 4.6 approach; correctness looks sound.
litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py	Correctly excludes Opus 4.7 from _supports_tool_search_on_bedrock with an explanatory comment; strips output_config for all Bedrock Invoke calls; extends extended-thinking model list to include Opus 4.7.
litellm/llms/bedrock/chat/converse_transformation.py	Silently drops tool_search_tool_* tools for all Converse models; adds Opus 4.7 to the computer-use-2025-11-24 beta-header list. Changes look correct.
model_prices_and_context_window.json	Adds Opus 4.7 entries for anthropic, bedrock (cross-region prefixes), vertex_ai, and azure_ai with correct pricing and capability flags; supports_xhigh_reasoning_effort: true is set. Missing supports_max_reasoning_effort (intentional — max is Opus 4.6-only).
tests/test_litellm/llms/bedrock/messages/invoke_transformations/test_anthropic_claude3_transformation.py	Adds mock-only tests for output_config stripping, SSE usage promotion, and clear_thinking injection; all tests are unit-level with no real network calls. No new Opus 4.7-specific test for tool-search exclusion.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User sends request with reasoning_effort or thinking] --> B{Model is Opus 4.7?}
    B -- Yes --> C[_map_reasoning_effort → type=adaptive]
    B -- No --> D[_map_reasoning_effort → type=enabled + budget_tokens]
    C --> E[Set output_config.effort via effort_map]
    E --> F{effort == max?}
    F -- Yes --> G[❌ ValueError: max is Opus 4.6 only]
    F -- No --> H{effort == xhigh?}
    H -- Yes --> I{_supports_effort_level via JSON?}
    I -- No --> J[❌ ValueError: xhigh not supported]
    I -- Yes --> K[Apply output_config to request]
    H -- No --> K
    K --> L{Provider?}
    L -- Anthropic Direct --> M[Send with output_config]
    L -- Bedrock Invoke --> N[Strip output_config step 5b]
    L -- Bedrock Converse --> O[Pass through normally]
    P[Tool search tools in request] --> Q{Provider?}
    Q -- Bedrock Converse --> R[Strip tool_search_tool_* for ALL models]
    Q -- Bedrock Invoke Opus 4.7 --> S[_supports_tool_search_on_bedrock → false]
    Q -- Bedrock Invoke Opus 4.5/4.6 --> T[Add tool-search-tool-2025-10-19 beta header]

Loading

_{Reviews (2): Last reviewed commit: "Remove max support for opus 4.7" | Re-trigger Greptile}

[ishaan-berri]

LGTM

[greptile-apps]

max effort check is hardcoded while xhigh is data-driven

effort='max' is validated with a hardcoded _is_opus_4_6_model() string match, while effort='xhigh' directly below uses _supports_effort_level(model, "xhigh") — exactly the pattern the project rule requires. The JSON schema already declares supports_max_reasoning_effort as a valid boolean field (see test_utils.py line 774). Any future model that supports max will require a code change here rather than a simple model-map addition.

The fix is a two-line change:

1. Add "supports_max_reasoning_effort": true to the claude-opus-4-6 (and its regional/versioned) JSON entries.
2. Replace the hardcoded gate:

Suggested change

	# ``max`` is Claude Opus 4.6 only (not Sonnet 4.6, not Opus 4.5/4.7).
	# Keep this hardcoded so the error message is specific and stable.
	if effort == "max" and not self._is_opus_4_6_model(model):
	raise ValueError(
	f"effort='max' is only supported by Claude Opus 4.6. Got model: {model}"
	f"effort='max' is only supported by Claude Opus 4.6. "
	f"Got model: {model}"
	)
	# ``xhigh`` is data-driven via ``supports_xhigh_reasoning_effort`` so
	# enabling it for a new model is a pure model-map change.
	if effort == "xhigh" and not self._supports_effort_level(model, "xhigh"):
	if effort == "max" and not self._supports_effort_level(model, "max"):
	raise ValueError(
	f"effort='max' is not supported by this model. Got model: {model}"
	)

Rule Used: What: Do not hardcode model-specific flags in the ... (source)