chore(release): backport Fable 5, batch-file auth, CrowdStrike AIDR, Mantle Responses SigV4, and NetApp streaming-cost fix to stable/1.88.x and cut 1.88.2

[mateo-berri]

Relevant issues

Backports to stable/1.88.x and bumps the patch version to 1.88.2 for a release:

• Add Claude Fable 5 across Anthropic, Bedrock, Vertex AI, and Azure AI #30064 — Claude Fable 5 across Anthropic, Bedrock, Vertex AI, and Azure AI (unchanged from the original revision of this PR)
• fix(proxy): authorize batch files using upload target_model_names (LIT-3593) #30009 — fix(proxy): authorize batch files using upload target_model_names (LIT-3593)
• CrowdStrike AIDR identity capture (feat(guardrails): capture user and model metadata in CrowdStrike AIDR #29517 feature, picked from stable/1.84.x as on the 1.85/1.86 backports) + fix(guardrails): read CrowdStrike AIDR identity from both metadata bags #29991 follow-up fix reading identity from both metadata bags
• feat(bedrock_mantle): add SigV4/IAM auth to Responses API route (fixes #29665) #29788 — feat(bedrock_mantle): add SigV4/IAM auth to the Responses API route, with its prerequisite feat(bedrock_mantle): add Responses API support (/openai/v1/responses) #29490 (the Mantle Responses route itself), which this line predates
• fix: completion_cost AttributeError on streaming Anthropic web_search responses (#26153) #27346 — fix: completion_cost AttributeError on streaming Anthropic web_search responses ([Bug]: completion_cost() raises AttributeError on streaming Anthropic responses with web_search in v1.83.10 #26153)

What is included

Commits in order:

• Add Claude Fable 5 across Anthropic, Bedrock, Vertex AI, and Azure AI #30064 Fable 5 (as previously reviewed on this PR)
• fix(proxy): authorize batch files using upload target_model_names (LIT-3593) #30009 batch-file auth, cherry-picked from staging 2cd7e87485 — applied cleanly
• CrowdStrike pair: feature pick 6fc715c5bd from stable/1.84.x + fix(guardrails): read CrowdStrike AIDR identity from both metadata bags #29991 staging squash 1bbaf1c39d
• feat(bedrock_mantle): add Responses API support (/openai/v1/responses) #29490 Mantle Responses route + feat(bedrock_mantle): add SigV4/IAM auth to Responses API route (fixes #29665) #29788 SigV4: both landed upstream only inside squashed OSS staging syncs (cb041966bf / 32c88ca74f), so each is applied as the exact squash diff of its PR head, attributed to the original author
• fix: completion_cost AttributeError on streaming Anthropic web_search responses (#26153) #27346 NetApp streaming-cost fix, cherry-picked from staging squash 4a3860df1f
• bump 1.88.1 → 1.88.2 and the matching uv.lock refresh

Conflict resolutions / branch adaptations

• CrowdStrike feature pick: same two conflicts as on 1.87.x (chore(release): backport Fable 5, batch-file auth, CrowdStrike AIDR, and Mantle Responses SigV4 to stable/1.87.x and cut 1.87.2 #30146) — kept this line's imports and guard_input.model_dump(mode="json"), adopted the dict[str, Any] annotation; appended the three feature tests then fix(guardrails): read CrowdStrike AIDR identity from both metadata bags #29991's parametrized test to this branch's test-file tail
• Formatting-only divergence from staging: fix(guardrails): read CrowdStrike AIDR identity from both metadata bags #29991's parametrized test block carries >88-col lines upstream; included here pre-formatted for this branch's black (same as chore(release): backport Fable 5, batch-file auth, CrowdStrike AIDR, and Mantle Responses SigV4 to stable/1.87.x and cut 1.87.2 #30146)
• feat(bedrock_mantle): add Responses API support (/openai/v1/responses) #29490 / feat(bedrock_mantle): add SigV4/IAM auth to Responses API route (fixes #29665) #29788: applied 100% cleanly (all hunks)
• fix: completion_cost AttributeError on streaming Anthropic web_search responses (#26153) #27346 NetApp: one conflict in llm_cost_calc/utils.py (the two new _get_token_detail_value / _get_web_search_requests helpers — kept the incoming block). Additionally, this line predates the Usage.__init__ server_tool_use dict→ServerToolUse coercion that staging carries (it landed via the squashed OSS sync Litellm oss staging 080626 #29932 / 32c88ca74f, not as a standalone commit). calculate_usage round-trips the usage through Usage(**returned_usage.model_dump()), which re-serializes server_tool_use to a plain dict — so without that coercion the rebuilt usage holds a dict and the regression test asserting a ServerToolUse type fails. Restored the coercion in litellm/types/utils.py; this matches fix: completion_cost AttributeError on streaming Anthropic web_search responses (#26153) #27346's own first commit ("coerce server_tool_use dict to ServerToolUse in Usage.init"), dropped from the squash only because staging already had it.

Verified on this branch

• Picked-change suites: test_batch_file_validation.py, test_crowdstrike_aidr.py, test_bedrock_mantle_responses_transformation.py, test_llm_http_handler.py — 114 passed
• NetApp suites: test_tool_call_cost_tracking_dict_safety.py, test_streaming_chunk_builder_server_tool_use.py, test_streaming_chunk_builder_utils.py, test_cost_calculation_dict_safety.py — 27 passed; broader litellm_core_utils usage/cost sweep 115 passed
• Fable suites previously verified (18+21+2 passed, unchanged)
• Both model-map JSONs parse; black clean on all hand-touched files

Pre-Submission checklist

• The cherry-picked PRs carry their own tests
• Scope is limited to backporting already-merged changes plus the release bump

Type

🐛 Bug Fix
🆕 New Feature (backport)

[codecov]

Codecov Report

❌ Patch coverage is 14.28571% with 114 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
litellm/llms/anthropic/common_utils.py	24.44%	34 Missing ⚠️
...ardrail_hooks/crowdstrike_aidr/crowdstrike_aidr.py	4.54%	21 Missing ⚠️
litellm/llms/custom_httpx/llm_http_handler.py	0.00%	16 Missing ⚠️
litellm/litellm_core_utils/llm_cost_calc/utils.py	25.00%	9 Missing ⚠️
...tellm/llms/bedrock/chat/converse_transformation.py	14.28%	6 Missing ⚠️
litellm/proxy/hooks/batch_rate_limiter.py	0.00%	6 Missing ⚠️
...itellm_core_utils/streaming_chunk_builder_utils.py	0.00%	5 Missing ⚠️
litellm/llms/anthropic/chat/transformation.py	0.00%	5 Missing ⚠️
litellm/llms/anthropic/cost_calculation.py	0.00%	5 Missing ⚠️
litellm/utils.py	0.00%	5 Missing ⚠️
... and 2 more

📢 Thoughts on this report? Let us know!

[greptile-apps]

Greptile Summary

Backports Claude Fable 5 support from main to stable/1.88.x, adding model-map entries across Anthropic, Bedrock (base + global/us/eu profiles), Vertex AI (@default), and Azure AI, together with sampling-params gating (supports_sampling_params: false) shared by Fable 5 and the already-landed Opus 4.7/4.8 models.

• Adds _apply_sampling_param / _supports_sampling_params helpers to AnthropicModelInfo that read the new supports_sampling_params flag from the cost map and either drop or raise a clean UnsupportedParamsError based on drop_params; the gate is applied at both the map_openai_params and transform_request boundaries in the Anthropic and Bedrock Converse paths.
• Fixes the top_k=0 truthiness bug in _handle_top_k_value (if val_top_k: → if val_top_k is not None:) so zero is no longer silently swallowed.
• Ships 41 new mock tests covering drop/raise behaviour, model-map-flag overrides, temperature=1 passthrough, and per-provider top_k routing.

Confidence Score: 4/5

Safe to merge; all model entries, gating logic, and drop/raise wiring are correct for currently supported models.

The implementation is well-structured and the test coverage is thorough. The one issue is in _supports_sampling_params: after the cost-map lookup returns None (model not found), the fallback falls back to a hardcoded list of model-name substrings rather than defaulting to True. This is redundant for every model this PR supports (they all carry the flag in the map) and violates the team rule against in-code model flags; a future model sharing any of those substrings could be incorrectly gated.

litellm/llms/anthropic/common_utils.py — the name-based fallback in _supports_sampling_params (lines 288-302) is the only area worth a second look.

Important Files Changed

Filename	Overview
litellm/llms/anthropic/common_utils.py	Adds sampling-params gating helpers (_supports_sampling_params, _apply_sampling_param, _model_map_lookup_candidates, _get_model_capability, _supports_model_capability) and refactors _is_adaptive_thinking_model; the primary cost-map-driven path is correct but the hardcoded name-based fallback in _supports_sampling_params violates the team rule against in-code model flags.
litellm/llms/anthropic/chat/transformation.py	Routes temperature/top_p through _apply_sampling_param in map_openai_params, and gates top_k at the transform_request boundary; logic and drop_params wiring look correct.
litellm/llms/bedrock/chat/converse_transformation.py	Propagates drop_params through _prepare_request_params and _handle_top_k_value, fixes top_k=0 truthiness bug (if val_top_k → if val_top_k is not None), and delegates sampling-param gating to _apply_sampling_param; all changes look correct.
model_prices_and_context_window.json	Adds 8 Fable 5 entries (anthropic, bedrock base + global/us/eu, vertex_ai + @default, azure_ai) and backfills supports_sampling_params: false on existing opus-4-7/4-8 entries; all flags and pricing look consistent.
tests/test_litellm/test_claude_fable_5_config.py	New test file with 18 mock/unit tests covering pricing, capability flags, adaptive-thinking detection, sampling-param flag presence, BEDROCK_CONVERSE_MODELS registration, and backup parity; all network-free and appropriate for this directory.

_{Reviews (1): Last reviewed commit: "Add Claude Fable 5 across Anthropic, Bed..." | Re-trigger Greptile}

[greptile-apps]

Hardcoded model-name fallback violates team rule

The name-based fallback ("fable", "opus-4-7", …) hardcodes which models reject sampling params directly in Python, which the team rule explicitly prohibits. All currently affected models now carry supports_sampling_params: false in the cost map, and _model_map_lookup_candidates already strips provider prefixes (bedrock/, bedrock/converse/, vertex_ai/, etc.), so every valid provider-routed ID resolves to a flagged entry without this fallback. The residual risk is that a future model whose name contains "fable" or "opus-4-7" but does support sampling params would silently be gated incorrectly. Remove the string-matching block and let an absent flag default to True (supports sampling params), which is the safe assumption.

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

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
5 out of 6 committers have signed the CLA.

✅ yuneng-berri
✅ kenany
✅ kingdoooo
✅ mateo-berri
✅ Sameerlite
❌ ishaan-berri
_{You have signed the CLA already but the status is still pending? Let us recheck it.}