●Change8

litellm-trace-dev-v1.81.16

📦 litellmView on GitHub →
✨ 24 features🐛 20 fixes⚡ 1 deprecations🔧 14 symbols

Summary

This release introduces significant feature enhancements, particularly around UI projects, access control, and model support, alongside numerous bug fixes and performance optimizations across various integrations and internal systems.

Migration Steps

  1. If using older xAI models grok-2-vision-1212 or grok-3-mini, note their deprecation dates and plan migration.

✨ New Features

  • Added support for gemini-3.1-flash-image-preview model on Vertex AI.
  • Added v1 transformation for Anthropic responses.
  • Enabled local file support for OCR.
  • Implemented Access group CRUD with bidirectional team/key synchronization.
  • Added gpt-audio-1.5 and gpt-realtime-1.5 models to the cost map.
  • Added OpenRouter native models to the model cost map.
  • Included timestamps in the /project/list response.
  • Added UI for Projects, including list and create flows.
  • Added Prometheus child_exit cleanup for gunicorn workers.
  • Added in_flight_requests metric to /health/backlog endpoint and Prometheus.
  • Added Agent RBAC Permission fix to ensure internal users cannot create agents.
  • Added PROXY_ADMIN role to system user for key rotation.
  • Populated user_id and user_info for admin users in /user/info endpoint.
  • Propagated extra_headers to Upstream for image generation requests.
  • Shortened guardrail benchmark result filenames for Windows long path support.
  • Set default available_on_public_internet to true in MCP configuration.
  • Improved streaming latency with 4 targeted hot-path fixes.
  • Added Project ID and access_group_id filters to the Key list endpoint.
  • Added Project Details Page to the UI.
  • Added project keys table and project dropdown to key create/edit flows in the UI.
  • Implemented Litellm health check tokens.
  • Added ability to trace metrics to Datadog.
  • Added duplicate issue detection and auto-close bot to CI.
  • Implemented Litellm presidio stream v3.

🐛 Bug Fixes

  • Fixed converse handling for parallel_tool_calls.
  • Preserved forwarding server side called tools.
  • Fixed free models working correctly from the UI.
  • Added ChatCompletionImageObject in OpenAIChatCompletionAssistantMessage.
  • Fixed poetry lock issues.
  • Stripped stale mcp-session-id to prevent 400 errors across proxy workers.
  • Fixed function calling for PublicAI Apertus models.
  • Fixed Claude code plugin schema.
  • Added missing migration for LiteLLM_ClaudeCodePluginTable.
  • Restored parallel_tool_calls mapping in map_openai_params for Bedrock.
  • Updated test mocks for renamed filter_server_ids_by_ip_with_info.
  • Fixed caching by storing task references in LLMClientCache._remove_key.
  • Passed MCP auth headers from request into tool fetch for /v1/responses and chat completions.
  • Fixed filtering of internal json_tool_call when mixed with real tools in Bedrock.
  • Fixed OIDC discovery URLs, roles array handling, and dot-notation error hints in JWT handling.
  • Updated realtime guardrail test assertions for voice violation behavior.
  • Updated Azure pass-through test after Responses API routing change.
  • Fixed guardrail benchmark result filenames for Windows long path support.
  • Fixed MCP configuration by defaulting available_on_public_internet to true.
  • Fixed JWT handling issues related to OIDC discovery URLs, roles array, and dot-notation errors.

Affected Symbols

gemini-3.1-flash-image-previewgpt-audio-1.5gpt-realtime-1.5grok-2-vision-1212grok-3-miniOpenRouter Opus 4.6Claude Opus 4.6AssemblyAI Universal-3 ProAssemblyAI Speech UnderstandingAssemblyAI LLM GatewayLLMClientCacheChatCompletionImageObjectOpenAIChatCompletionAssistantMessageLiteLLM_ClaudeCodePluginTable

⚡ Deprecations

  • Added deprecation dates for xAI grok-2-vision-1212 and grok-3-mini models.