●Change8

v1.82.2-silent.dev

📦 litellmView on GitHub →
✨ 20 features🐛 24 fixes🔧 22 symbols

Summary

This release focuses heavily on enhancing Multi-Cloud Proxy (MCP) functionality, including new authentication methods, BYOM support, and UI improvements. It also brings numerous bug fixes across providers like Gemini, Anthropic, and OpenAI, alongside new feature support for transcription and image editing.

Migration Steps

  1. Migrate user management buttons from Tremor to Ant Design (UI change, may require frontend adjustments if interacting directly with old components).
  2. If using Anthropic tools, ensure tool input schemas enforce type:"object" where necessary, as fixes were applied to _map_tool_helper.

✨ New Features

  • Auto-inject prompt caching support for Gemini models.
  • Support detail parameter for image resolution on Gemini 2.x models.
  • Map reasoning to reasoning_content in Delta for gpt-oss providers during streaming.
  • Add Voxtral audio transcription support for Mistral.
  • Add litellm.acount_tokens() public API and OpenAI token counting support.
  • Add image generation and image edit support for black_forest_labs.
  • Add token authentication support for MCP servers.
  • Implement PANW Prisma AIRS guardrail with apply_guardrail support.
  • Introduce policy builder functionality (though later reverted, keeping track of intent).
  • Add support for BYOM (Bring Your Own Model) - non-admin MCP server submission and admin review workflow.
  • Implement OpenAPI MCP server support with popular API quick-picker in UI.
  • Add missing supports_web_search flag to OpenAI models.
  • Support input_fidelity parameter for image edit API.
  • Add model_cost aliases expansion support.
  • Implement Responses API support for JSON providers (openai_like).
  • Add support for qwen3.5 series for openrouter.
  • Add support for gemini embedding 2 preview.
  • Implement team bring-your-own guardrails.
  • Add MCP tool execution in /chat via responses API.
  • Implement UI features: MCP OAuth2 value prop/auth badges, show MCP tools per server in chat panel, show logos on MCP Apps page, Per-server health recheck for MCP Servers.

🐛 Bug Fixes

  • Avoid migration hook serviceaccount dependency cycle.
  • Set budget_reset_at when creating customer with budget_duration.
  • Fix user specified async client being ignored with Gemini streaming+async.
  • Add missing indexes for top CPU-consuming queries.
  • Fix Claude code req traces on langfuse.
  • Fix Presidio edge case with Anthropic handle on PII token leak.
  • Map unknown finish_reason values to finish_reason_unspecified in stream_chunk_builder to prevent ValidationError during streaming.
  • Strip secret_fields from guardrail logging response for security.
  • Fix Anthropic prompt caching returning a string instead of an int.
  • Fix tag usage cost panel showing zero.
  • Fix MCP: don't auto-detect M2M OAuth from field presence.
  • Fix MCP: add AWS SigV4 auth for Bedrock AgentCore MCP servers.
  • Fix Snowflake: transform tool_choice string to object format.
  • Fix Vertex: skip schema transforms for Gemini 2.0+ tool parameters.
  • Preserve thinking block order with multiple web searches.
  • Unify finish_reason mapping to OpenAI-compatible values during completion.
  • Fix MCP OpenAPI OAuth flow issues (transport mapping, error messages, discovery bypass).
  • Fix UI: stop health checks triggering on MCP server deletion.
  • Fix MCP Key Scope Type.
  • Fix Anthropic: deduplicate tool_result messages by tool_call_id.
  • Fix Anthropic: enforce type:"object" on tool input schemas in _map_tool_helper.
  • Fix (xai): streaming empty chunk bug for providers using BaseLLMHTTPHandler.
  • Fix proxy: preserve multipart/form-data boundary in passthrough endpoints.
  • Fix proxy startup failing if prisma migrate fails.

Affected Symbols

litellm.acount_tokens()Gemini 2.x modelsgpt-oss providersVoxtralblack_forest_labsGemini streaming+async client handlingClaude.mdpresidio/anthropic PII handlingstream_chunk_builderguardrail loggingMCP serversAnthropic prompt cachingBedrock AgentCore MCP serversSnowflake tool_choice handlingVertex Gemini 2.0+ tool parametersOpenAI models (supports_web_search)image edit APIgpt-5.4 (reasoning_effort, tools limitation, tool_search, namespaces)OpenAI-compatible finish_reason mappingqwen3.5gemini embedding 2 previewBaseLLMHTTPHandler