●Change8

v1.79.1.dev1

📦 litellmView on GitHub →
✨ 21 features🐛 22 fixes⚡ 1 deprecations🔧 14 symbols

Summary

This release introduces significant enhancements across observability (OpenTelemetry), guardrails, and the UI, alongside numerous bug fixes for providers like Azure OpenAI and improved performance.

✨ New Features

  • Upgraded Lasso API to v3 and fixed ULID generation.
  • Enabled OpenTelemetry context propagation by external tracers.
  • Added support for dynamic client registration.
  • Support for `during_call` for model armor guardrails.
  • Added `base_url` config with environment variables support for OpenRouter.
  • Added support for logging error information on OTEL.
  • UI: Changed API Base from Select to Input in New LLM Credentials.
  • Added /openai routes for Responses API (Azure OpenAI SDK Compatibility).
  • Added GitlabPromptCache and enabled subfolder access.
  • Added Haiku 4.5 pricing for open router.
  • Graceful degradation for pillar service when using litellm.
  • Python entry-point for CustomLLM subclasses.
  • Allow using ARNs when generation images via Bedrock.
  • Added FAL AI Image Generations on LiteLLM.
  • Added Mistral API model codestral-embed-2505.
  • Added support for Batch API Rate limiting (input based rate limits).
  • UI: Validation for Proxy Base URL in SSO Settings.
  • UI: Test Key UI Embeddings.
  • UI: Add Key Type Select in Key Settings.
  • Per-request profile overrides added to PANW Prisma AIRS guardrails.
  • Added OpenAI-compatible annotations support for Cohere v2 citations.

🐛 Bug Fixes

  • Fixed duplicate trace in langfuse_otel.
  • Updated IBM Guardrails to correctly use SSL Verify argument.
  • Azure OpenAI: Fixed ContextWindowExceededError not being mapped from Azure openai errors.
  • DD logging: Ensured key's metadata + guardrail is logged on DD.
  • Minor fix proxy: Ensured User API key and team id and user id missing from custom callback is not misfiring.
  • Azure OpenAI: Added handling for `v1` under azure api versions.
  • Fixed respecting `LiteLLM-Disable-Message-Redaction` header for Responses API.
  • Removed limit from admin UI numerical input fix.
  • UI: Added Key Already Exist Error Notification.
  • Preserved Bedrock inference profile IDs in health checks.
  • Fixed support for tool usage messages with Langfuse OTEL integration.
  • Enhanced requester metadata retrieval from API key auth for opik.
  • Fixed mutation of original request for gemini request.
  • Fixed redacting reasoning summaries in ResponsesAPI output when message logging is disabled.
  • Fixed support for text.format parameter in Responses API for providers without native ResponsesAPIConfig.
  • Removed unnecessary model variable assignment.
  • Prevented memory leaks from jitter and frequent job intervals in apscheduler.
  • Added fallback logic for detecting file content-type when S3 returns generic.
  • Prevented httpx DeprecationWarning memory leak in AsyncHTTPHandler.
  • Fixed spend tracking for OCR/aOCR requests (log `pages_processed` + recognize `OCRResponse`).
  • Fixed Circle CI UI E2E Test.
  • Moderations endpoint now respects `api_base` configuration parameter.

🔧 Affected Symbols

Lasso API v3OpenTelemetryIBM GuardrailsAzure OpenAIDD loggingLangfuse OTELResponses APIBedrockOpikGitlabPromptCacheGeminiCohere v2PANW Prisma AIRShttpx.AsyncHTTPHandler

⚡ Deprecations

  • Added deprecation dates for models.