fix(bedrock): drop toolSpec.strict for Claude Sonnet 5 and Fable 5

[seanturner83]

Relevant issues

Follow-up to #31923 (Opus 4.7/4.8) and #31943 (Sonnet 4) — same root cause, two more models.

Pre-Submission checklist

• I have added meaningful tests
• My PR passes all CI/CD checks (lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem

What this fixes

Bedrock Converse routes Claude Sonnet 5 and Fable 5 through the same Anthropic-compatible validator as Opus 4.7/4.8 and Sonnet 4, which rejects toolSpec.strict even though Anthropic's native API accepts strict as a top-level tool field:

tools.0.custom.strict: Extra inputs are not permitted

Both models were missing the bedrock_converse_supports_strict_tools: false opt-out flag, so the default strict-forwarding path 400s any tool-calling request against bedrock/*.claude-sonnet-5 or bedrock/*.claude-fable-5 (SDKs like openai-agents stamp strict on every tool, so it fires universally at the first call).

The change

• Add bedrock_converse_supports_strict_tools: false to all bedrock_converse Sonnet 5 (6) and Fable 5 (4) inference-profile entries in model_prices_and_context_window.json and its backup.
• Extend the existing strict-tools regression tests (test_bedrock_converse_strict_tools_opus_47_48.py) to cover Sonnet 5 / Fable 5 in the dropped-list, the helper check, and the cost-map-flag check.

No logic change — this reuses the exact mechanism from #31923/#31943; the models just needed the flag.

Screenshots / Proof of Fix

Live Bedrock Converse calls (us-east-1, real $), captured at commit bbb8af5:

# us.anthropic.claude-sonnet-5
  BEFORE (strict forwarded, as litellm does today): 400 ValidationException — tools.0.custom.strict: Extra inputs are not permitted
  AFTER  (strict stripped, what this flag does):     200 OK (stopReason=end_turn)

# us.anthropic.claude-fable-5
  BEFORE (strict forwarded, as litellm does today): 400 ValidationException — tools.0.custom.strict: Extra inputs are not permitted
  AFTER  (strict stripped, what this flag does):     200 OK (stopReason=end_turn)

Confirmed for both us. and global. inference-profile variants of each model. Unit tests: pytest tests/test_litellm/litellm_core_utils/prompt_templates/test_bedrock_converse_strict_tools_opus_47_48.py → 44 passed.

Type

🐛 Bug Fix

Co-authored with Claude Opus 4.8.

[greptile-apps]

Greptile Summary

This PR adds bedrock_converse_supports_strict_tools: false to all Bedrock inference-profile entries for Claude Sonnet 5 (6 entries) and Claude Fable 5 (4 entries) in both the primary and backup model-pricing JSON files, preventing toolSpec.strict from being forwarded to Bedrock Converse where it causes a ValidationException. Tests are extended to cover the new models in all three test functions.

• JSON data change only: no logic modified; follows the identical pattern established by fix(bedrock/converse): drop toolSpec.strict for Opus 4.7/4.8 (#31582) #31923 and fix(bedrock): drop strict/additionalProperties from toolSpec for Claude Sonnet 4 #31943, and both JSON files are kept in sync.
• Tests expanded: all 10 new cost-map keys are covered in test_strict_tools_flag_set_in_model_cost_map, the full-pipeline _dropped test covers all 10 model IDs (6 Sonnet 5 + 4 Fable 5 with and without bedrock/ prefix), and the helper function test now asserts False for both bare and prefixed Sonnet 5 / Fable 5 forms.

Confidence Score: 5/5

Safe to merge — the change is additive JSON data only, touching no runtime logic, and every new entry is covered by tests.

The fix follows an established, tested pattern from two prior PRs. Both JSON files are kept in sync, all 10 new cost-map keys have direct test coverage across three test functions, and the live-call evidence in the PR description confirms the before/after behaviour on real Bedrock endpoints.

No files require special attention.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds bedrock_converse_supports_strict_tools: false to 10 model entries (6 Sonnet 5, 4 Fable 5); consistent with existing flag placement for Opus 4.7/4.8 and Sonnet 4.
litellm/model_prices_and_context_window_backup.json	Identical flag additions as the primary JSON; the two files are kept in sync.
tests/test_litellm/litellm_core_utils/prompt_templates/test_bedrock_converse_strict_tools_opus_47_48.py	Extends parametrized tests for all three test functions to cover Sonnet 5 and Fable 5; function docstring updated; module-level docstring still only mentions Opus 4.7/4.8 and Sonnet 4.

_{Reviews (2): Last reviewed commit: "test(bedrock): cover all Sonnet 5 / Fabl..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will improve performance by 13.29%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
✅ 29 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	test_completion_simple_message	4.6 ms	4.1 ms	+13.29%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing seanturner83:fix/bedrock-converse-drop-strict-sonnet5-fable5 (70ee6ff) with litellm_internal_staging (cd6e8cd)}

[seanturner83]

Thanks @greptileai — addressed both gaps in 70ee6ff:

• Added the 4 missing regional variants to test_bedrock_tools_pt_strict_dropped_for_strict_unsupported_models (the full-pipeline test): au./jp. Sonnet 5 and global./eu. Fable 5.
• Added the bare anthropic.claude-fable-5 key to test_bedrock_converse_supports_strict_tools_helper.
• Fixed the stale docstring to include Sonnet 5 and Fable 5.

All 10 cost-map keys now have end-to-end _bedrock_tools_pt coverage. Tests: 48 passed.

@greptileai