Align Claude Opus 4.6 metadata and limits

[PeterDaveHello]

Relevant issues

#20506 #20508

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
  • Updated: tests/test_litellm/test_claude_opus_4_6_config.py
• My PR passes all unit tests on make test-unit
  • Ran targeted tests for this change:
    • poetry run pytest tests/test_litellm/test_claude_opus_4_6_config.py -q
    • poetry run pytest tests/test_litellm/test_utils.py::test_aaamodel_prices_and_context_window_json_is_valid tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py::test_get_supported_openai_params_bedrock_converse -q
  • Results:
    • 4 passed in 0.18s
    • 2 passed in 0.59s
• My PR's scope is as isolated as possible, it only solves 1 specific problem

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

Commits in this PR (main..HEAD)

• 7f5116c0ec Align Claude Opus 4.6 metadata and limits
• ee53bcdf76 Add Bedrock v1 aliases for Claude Opus 4.6

Changes

1. Align Claude Opus 4.6 metadata and long-context limits

   • Align claude-opus-4-6 and claude-opus-4-6-20260205 capability metadata:
     • supports_assistant_prefill=false
     • tool_use_system_prompt_tokens=346
   • Set max_input_tokens=1000000 for Opus 4.6 entries that include *_above_200k_tokens pricing:
     • Anthropic alias + dated ID
     • Bedrock Converse :0 base/global/regional IDs
     • Vertex AI Opus 4.6
   • Keep Azure AI Opus 4.6 at max_input_tokens=200000 (no long-context pricing fields configured)

2. Add Bedrock Converse Opus 4.6 aliases without :0

   • Add explicit pricing entries for:
     • anthropic.claude-opus-4-6-v1
     • global.anthropic.claude-opus-4-6-v1
     • us.anthropic.claude-opus-4-6-v1
     • eu.anthropic.claude-opus-4-6-v1
     • apac.anthropic.claude-opus-4-6-v1
   • Preserve global vs regional pricing split (regional entries include the 10% premium)
   • Mirror all entries in both:
     • model_prices_and_context_window.json
     • litellm/model_prices_and_context_window_backup.json

3. Extend regression coverage

   • Validate alias vs dated Opus 4.6 metadata parity
   • Validate Bedrock global/regional pricing for both :0 and non-:0 IDs
   • Validate Bedrock Converse model registration for Opus 4.6 IDs

Notes

• This is a follow-up PR on top of the initial Opus 4.6 model addition, focused on consistency and routing/cost-map correctness.
• The scope is intentionally limited to Opus 4.6 mappings and related tests only.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 6, 2026 6:42am

[greptile-apps]

Greptile Overview

Greptile Summary

Aligns Claude Opus 4.6 model metadata and adds Bedrock Converse aliases without version suffix for improved routing flexibility.

Changes include updating token limits from 200k to 1M for all Opus 4.6 entries with long-context pricing, aligning metadata across alias and dated models, adding 5 new Bedrock Converse aliases (base, global, regional variants), preserving regional pricing premiums, and expanding test coverage for pricing parity and model registration. Azure AI remains at 200k tokens due to lack of long-context pricing configuration.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes are purely configuration updates to model pricing and metadata, fully covered by tests. Regional pricing calculations are mathematically correct (10% premium verified). Both JSON files are identical. The new models will be automatically registered via existing infrastructure. No code logic changes, no breaking changes.
• No files require special attention

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds Bedrock aliases for Opus 4.6, aligns max tokens to 1M for long-context variants, updates metadata
litellm/model_prices_and_context_window_backup.json	Mirrors the main JSON file changes identically
tests/test_litellm/test_claude_opus_4_6_config.py	Expands test coverage for non-:0 Bedrock aliases, validates pricing parity, adds metadata consistency test between alias and dated model

Sequence Diagram

sequenceDiagram
    participant User
    participant LiteLLM
    participant ModelCostMap
    participant BedrockConverse
    
    Note over User,BedrockConverse: PR Changes: Add non-:0 Bedrock aliases for Claude Opus 4.6
    
    User->>LiteLLM: Request with model="anthropic.claude-opus-4-6-v1"
    LiteLLM->>ModelCostMap: Load model_prices_and_context_window.json
    ModelCostMap-->>LiteLLM: Model config (litellm_provider: bedrock_converse)
    
    Note over ModelCostMap: New entries added:<br/>- anthropic.claude-opus-4-6-v1<br/>- global.anthropic.claude-opus-4-6-v1<br/>- us/eu/apac.anthropic.claude-opus-4-6-v1
    
    LiteLLM->>LiteLLM: add_known_models() runs
    Note over LiteLLM: Automatically adds models with<br/>litellm_provider="bedrock_converse"<br/>to bedrock_converse_models set
    
    LiteLLM->>BedrockConverse: Route to Bedrock Converse API
    Note over BedrockConverse: Uses aligned metadata:<br/>- max_input_tokens: 1M<br/>- supports_assistant_prefill: false<br/>- tool_use_system_prompt_tokens: 346
    
    BedrockConverse-->>LiteLLM: Response
    LiteLLM->>LiteLLM: Calculate cost with regional pricing
    Note over LiteLLM: Regional pricing (10% premium):<br/>us/eu/apac vs global
    
    LiteLLM-->>User: Response with accurate cost

Loading

[greptile-apps]

_{3 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[marty-sullivan]

@PeterDaveHello Thanks for jumping on this. AWS has also updated their pricing page to include Opus 4.6. I do see there are pricing differences between the regional endpoint and the inference profiles, so do make sure these are also aligned:

https://aws.amazon.com/bedrock/pricing/

[marty-sullivan]

Also, I assume the litellm/constants.py for bedrock converse models also needs to be aligned with

anthropic.claude-opus-4-6-v1

instead of:

anthropic.claude-opus-4-6-v1:0

[ishaan-jaff]

@PeterDaveHello please address the comment above

[ishaan-jaff]

bump @PeterDaveHello