What's Changed
- [Issue#21344]: avoid migration hook serviceaccount dependency cycle by @devarakondasrikanth in #21405
- Bug Fix: auto-inject prompt caching support for Gemini models by @RheagalFire in #21881
- fix: set budget_reset_at when creating customer with budget_duration by @araman-godaddy in #22009
- fix(gemini): support detail parameter for image resolution on Gemini 2.x models by @Chesars in #17662
- fix(streaming): map reasoning to reasoning_content in Delta for gpt-oss providers by @Chesars in #22803
- feat(mistral): add Voxtral audio transcription support by @Chesars in #22801
- Fix model management page UI improvements by @Chesars in #19518
- Fix: User specified async client ignored with Gemini streaming+async by @Chesars in #17550
- feat(openai): add litellm.acount_tokens() public API + OpenAI token counting support by @Chesars in #22809
- feat(black_forest_labs): add image generation and image edit support by @Chesars in #18144
- docs: add DB query performance guidelines to CLAUDE.md by @CAFxX in #23196
- fix: add missing indexes for top CPU-consuming queries by @CAFxX in #23147
- fix: claude code req traces on langfuse by @Harshit28j in #23259
- fix: presidio edge case with antropic handle on pii token leak by @Harshit28j in #22627
- fix(streaming): map unknown finish_reason values to finish_reason_unspecified to prevent ValidationError in stream_chunk_builder by @xykong in #22673
- fix(security): strip secret_fields from guardrail logging response by @milan-berri in #23162
- feat(mcp): add token authentication support for MCP servers by @milan-berri in #23260
- Migrate user management buttons from Tremor to Ant Design by @ryan-crabbe in #23215
- fix: anthropic prompt caching returns a string instead of an int by @mubashir1osmani in #23229
- fix: tag usage cost panel zero by @mubashir1osmani in #23202
- fix(mcp): don't auto-detect M2M OAuth from field presence by @ishaan-jaff in #23187
- fix(mcp): add AWS SigV4 auth for Bedrock AgentCore MCP servers by @michelligabriele in #22782
- feat(panw-prisma-airs): PANW Prisma AIRS guardrail with apply_guardrail support by @jroberts2600 in #22999
- fix(snowflake): transform tool_choice string to object format by @stevejaker in #23268
- policy builder by @shivamrawat1 in #23287
- docs: pip venv upgrade workflow by @milan-berri in #23290
- docs(openai): document gpt-5.4 reasoning_effort + tools limitation by @Chesars in #23271
- feat(mcp): BYOM — non-admin MCP server submission + admin review workflow by @ishaan-jaff in #23205
- feat(ui): OpenAPI MCP server support with popular API quick-picker by @ishaan-jaff in #23200
- feat(openai): add missing supports_web_search to OpenAI models by @Chesars in #23275
- fix(vertex): skip schema transforms for Gemini 2.0+ tool parameters by @Chesars in #23265
- feat(images): support input_fidelity parameter for image edit API by @Chesars in #23201
- fix: preserve thinking block order with multiple web searches by @MaxwellCalkin in #23093
- feat: add model_cost aliases expansion support by @Chesars in #21601
- feat(chat-ui): add MCP OAuth2 value prop and server auth badges by @ishaan-jaff in #23291
- feat(ui): show MCP tools per server in chat panel by @ishaan-jaff in #23294
- docs(responses): add tool_search & namespaces docs for gpt-5.4 by @Chesars in #23258
- fix(completion): unify finish_reason mapping to OpenAI-compatible values by @Chesars in #22138
- Revert "policy builder" by @shivamrawat1 in #23295
- fix(mcp): fix OpenAPI OAuth flow — transport mapping, error messages, discovery bypass by @ishaan-jaff in #23300
- [Fix] UI - MCP Servers: stop health checks triggering on server deletion by @yuneng-jiang in #23063
- [Feature] UI - Keys: Improve VirtualKeysTable Column Display by @yuneng-jiang in #23301
- Team bring-your-own guardrails by @krrishdholakia in #23307
- Revert "feat: add model_cost aliases expansion support" by @Chesars in #23313
- feat(chat-ui): responses API + MCP tool execution in /chat by @ishaan-jaff in #23297
- feat: add model_cost aliases expansion support by @Chesars in #23314
- feat(model): add qwen3.5 series for openrouter by @FacerAin in #23135
- feat(openai_like): add Responses API support to JSON providers by @Chesars in #21398
- [Feat] UI - Show logos on MCP Apps page by @ishaan-jaff in #23320
- fix(codeql): remove ruby from language matrix by @joereyna in #23227
- [Feature] MCP Server Team-Scoped Filtering for Key Creation by @yuneng-jiang in #23323
- [Fix] MCP Key Scope Type Fix by @yuneng-jiang in #23326
- fix(anthropic): deduplicate tool_result messages by tool_call_id by @netbrah in #23104
- fix(anthropic): enforce type:"object" on tool input schemas in _map_tool_helper by @netbrah in #23103
- [Feature] UI - MCP Servers: Per-server health recheck by @yuneng-jiang in #23328
- [Test] UI - Logs: Add unit tests for view_logs components by @yuneng-jiang in #23298
- build(deps): bump hono from 4.10.6 to 4.12.7 in /litellm-js/spend-logs by @dependabot[bot] in #23312
- Litellm oss staging 03 04 2026 by @krrishdholakia in #23163
- Litellm oss staging 03 10 2026 by @RheagalFire in #23276
- Fix (xai): streaming empty chunk bug for providers using BaseLLMHTTPHandler by @Chesars in #17155
- docs(web_fetch): add newer Claude models to supported models list by @michelligabriele in #23251
- [Feat]: Add support for gemini embedding 2 preview by @Sameerlite in #23322
- fix(proxy): preserve multipart/form-data boundary in passthrough endpoints by @Sameerlite in #23338
- fix: fail proxy startup if prisma migrate fails by @Harshit28j in #23257
- fix(ui): import MCPEvent type into local scope in chat/types.ts by @joereyna in #23330
- fix(codeql): switch to security-extended to fix OOM failures by @joereyna in #23226
- [Docs] Fix "Page Not Found" link for Anthropic endpoint by @PeterDaveHello in #23349
- Replace broken Anthropic Skills API integration test with unit tests by @joereyna in #23325
- Clean up skills test: remove duplicate imports, parameterize mock HTTP by @RheagalFire in #23360
- feat(mcp): add AWS SigV4 auth support in UI and fix credential merge on edit by @michelligabriele in #23282
- feat: Add Anthropic Files API support by @Chesars in #16594
- [Fix] Null safe checks for MCP Hub Tab by @yuneng-jiang in #23362
- Add Abilty To Set minReadySeconds From values Files by @rjduffner in #23173
- chore: cleanup deprecated models from pricing JSON by @Chesars in #20795
- docs: clarify when to use openai/responses/ prefix for built-in tools by @Chesars in #17717
- feat(vertex_ai): route region for partner models and add GLM support by @Chesars in #19104
- fix: batch key queries in list_team to eliminate N+1 by @CAFxX in #23152
- fix: use group_by instead of find_many(distinct) in /tag/list to avoid full table scan by @CAFxX in #23136
- [infra] Fixing UI Build and Tests by @yuneng-jiang in #23374
- style: run black formatter on /litellm by @Chesars in #23373
- Add Retool Assist tutorial with LiteLLM Proxy to docs by @Astrodevil in #21952
- [Fix] Flaky MCP Server and AgentCore Streaming Tests by @yuneng-jiang in #23377
- refactor(files): add type aliases for provider parameters by @peterkc in #17724
- [Fix] TypeError in spend tracking when request has no API key by @yuneng-jiang in #23363
- [Feature] Support auto_redirect_ui_login_to_sso in config.yaml by @yuneng-jiang in #23367
- feat: adding a timezone picker to the usage page by @ryan-crabbe in #23174
- [Fix] Ruff Lint Errors: Unused Imports, Undefined Names, Statement Counts by @yuneng-jiang in #23387
- [Fix] UI - Usage: Memoize Expensive Computations in UsagePageView by @yuneng-jiang in #23379
- [Infra] Add Daily Internal Dev Staging Branch by @yuneng-jiang in #23394
- fix(mcp): fix OpenAPI OAuth flow — transport mapping, error messages, and discovery bypass by @ishaan-jaff in #23315
- [Fix] Update test_bad_database_url CI check for new error message by @yuneng-jiang in #23390
- Revert "feat: adding a timezone picker to the usage page" by @ryan-crabbe in #23399
- [Test] Add Unit Tests for 5 UI Components by @yuneng-jiang in #23393
- fix: fix default user perms respected through SCIM by @ryan-crabbe in #23395
- [Fix] Replace deprecated models in tests by @yuneng-jiang in #23400
- Revert "fix: add missing indexes for top CPU-consuming queries (#23147)" by @yuneng-jiang in #23402
- chore: bump proxy extras by @yuneng-jiang in #23404
- chore: regenerate poetry.lock to match pyproject.toml by @github-actions[bot] in #23405
- Fix 404 when fetching config-based MCP servers by ID by @shivamrawat1 in #22711
- Docs policyy builder by @shivamrawat1 in #23410
- v1.82.0 promote to stable by @shivamrawat1 in #23413
- fix: refetching mcp after creating by @ryan-crabbe in #23371
- Feat add token totals by @ryan-crabbe in #23181
- fix: null AWS SigV4 fields on MagicMock in TestTemporaryMCPSessionEndpoints by @joereyna in #23408
- fix: accept custom_body in URL-based create_pass_through_route endpoint_func by @joereyna in #23412
- style: make virtual keys tables' created by not a UUID by @ryan-crabbe in #23415
- LiteLLM CI stability by @yuneng-jiang in #23407
- tests: add ui unit tests for guardrail_info_helpers by @ryan-crabbe in #23422
- [Feature] Allow Admin Viewers to Access Audit Logs by @yuneng-jiang in #23419
- Revert "chore: cleanup deprecated models from pricing JSON" by @Sameerlite in #23426
- [Infra] CI/CD Fixes by @yuneng-jiang in #23425
- feat(ui): group MCP tools by CRUD risk category in allowlist panels by @ishaan-jaff in #23403
- fix(snowflake): transform string tool_choice to object format by @alvinttang in #23318
- fix(mcp): OAuth2 chat connect - tools fetch, auth, and status fixes by @ishaan-jaff in #23406
- fix: update stale gemini-1.5-flash model name in passthrough logging handler test by @joereyna in #23418
- fix: update stale model name in vertex AI batch cost calculation test by @joereyna in #23417
- Fix model cost for gemini-embedding-2-preview by @Sameerlite in #23431
- Litellm dev 03 10 2026 p1 by @krrishdholakia in #23429
- Litellm cicd 1203126 by @Sameerlite in #23448
- Add vector store retrieve list update delete by @Sameerlite in #23435
- fix(proxy): make async_post_call_response_headers_hook consistent across all endpoints by @michelligabriele in #22985
- fix: restore commits from revert of PR #21601 by @Chesars in #23457
- [Feat] Add WebRTC support by @Sameerlite in #23446
- feat(azure_ai): show actual model used in Azure Model Router response by @Sameerlite in #23432
- Improve doc for WebRTC by @Sameerlite in #23459
- Improve doc for WebRTC by @Sameerlite in #23460
- fix: check Black formatting in CI instead of auto-formatting by @Chesars in #18648
- [Infra] Fixing UI Build by @yuneng-jiang in #23463
- [Infra[ UI Build + Tests by @yuneng-jiang in https://github.com//pull/23465
- fix: strip SERVER_ROOT_PATH prefix before checking mapped pass-through routes by @joereyna in #23414
- Litellm oss staging 03 11 2026 by @RheagalFire in #23440
- fix(sso): direct PKCE token exchange + Redis wiring for multi-instance SSO by @ishaan-jaff in #22923
- [Infra] Proxy Extras by @yuneng-jiang in #23469
- Litellm ci stability by @yuneng-jiang in #23474
- [Fix] Remove recursion in BFL _read_image_bytes by @yuneng-jiang in #23477
- [Fix] Remove Deprecated o1-preview from Azure O-series Test by @yuneng-jiang in #23478
- [Fix] Ruff lint errors in MCP server files by @yuneng-jiang in #23482
- [Fix] Update Deprecated Model Names in CI Tests by @yuneng-jiang in #23483
- fix: add oauth2_flow="client_credentials" to MCPServer test helper by @joereyna in #23467
- [Fix] Add litellm-proxy-extras to CI requirements by @yuneng-jiang in #23486
- [Fix] Realtime websocket tests for websockets v15+ API by @yuneng-jiang in #23485
- [Fix] Ruff PLR0915 too-many-statements in ui_sso.py by @yuneng-jiang in #23487
- [Fix] Flaky and outdated router integration tests by @yuneng-jiang in #23488
- [Docs] Add PKCE_STRICT_CACHE_MISS to env variables reference by @yuneng-jiang in #23490
- Proxy fails to start on Windows with AttributeError: module 'os' has no attribute 'WNOHANG'. by @shivamrawat1 in #23494
- [Fix] BaseModelResponseIterator crashes on non-string stream chunks by @yuneng-jiang in #23497
- fix(gemini): preserve toolConfig on native generate_content by @emerzon in #23493
- [Fix] MyPy Errors by @yuneng-jiang in #23500
- QA: improve gpt-5.4 code/bugs by @Sameerlite in #23535
- Revert "QA: improve gpt-5.4 code/bugs" by
