●Change8

v1.80.10.dev.1

Breaking Changes
📦 litellmView on GitHub →
⚠ 1 breaking✨ 22 features🐛 24 fixes🔧 15 symbols

Summary

This release adds extensive new features—including Gemini 3 flash preview support, Azure Sentinel logging, Guardrails enhancements, and a new LinkUp Search provider—while delivering numerous bug fixes and a breaking change that renames `extra_headers` to `additional_headers`.

⚠️ Breaking Changes

  • Renamed parameter `extra_headers` to `additional_headers` in request handling; update code to use `additional_headers` instead of `extra_headers`.

Migration Steps

  1. Replace any usage of the `extra_headers` parameter with `additional_headers`.
  2. If using the Alpine Docker image on ARM64 for audio processing, pull the latest image which now includes libsndfile.

✨ New Features

  • Added Gemini 3 flash preview day 0 support.
  • Added SSO readiness health check endpoint.
  • Integrated Azure Sentinel logger.
  • Added new security scan to detect secret keys in the codebase.
  • Added diarize model for OpenAI.
  • Added secret manager settings controls to team management UI.
  • Implemented per‑team vault configuration.
  • Added email budget alerts.
  • Added UI cloud zero‑cost tracking.
  • Added Vertex AI Gemini 3 flash model to model map.
  • Added support for Bedrock image guardrails.
  • Added built‑in guardrails for harmful content, bias, etc.
  • Added content filter support for images in LiteLLM.
  • Added Lakera monitor mode.
  • Added new Search API provider – LinkUp Search.
  • Added reasoning support for vertex_ai/gemini‑3‑flash‑preview in pricing and context window.
  • Added gpt‑image‑1.5‑2025‑12‑16 to model cost map.
  • Added qwen 2 and qwen 3 models to Bedrock model ID mapping.
  • Added output_cost_per_image_token for gemini‑2.5‑flash‑image models.
  • Added support for Vertex multimodal embeddings (text + base64 image).
  • Added extraction of top‑level metadata for custom labels in Prometheus.
  • Added new guardrail option for OpenRouter models.

🐛 Bug Fixes

  • Corrected broken link in documentation.
  • Added libsndfile to Alpine Docker image for ARM64 audio processing.
  • Fixed get_model_from_request to correctly extract model ID from Vertex AI passthrough URLs.
  • Extracted top‑level metadata for custom labels in Prometheus.
  • Fixed guardrail handling for OpenRouter models and added a new guardrail option.
  • Returned AzureAnthropicConfig for Claude models in get_provider_chat_config.
  • Updated UI path handling for non‑root Docker deployments and restructured HTML.
  • Added drop_params support and fixed Vertex AI config for image_edit.
  • Prevented LiteLLM API key leakage on /health endpoint failures.
  • Added qwen 2 and qwen 3 handling in get_bedrock_model_id.
  • Fixed Vertex AI image edit credential source.
  • Fixed image URL handling.
  • Removed fake keys that caused security scans to fail.
  • Fixed /health/test_connection to resolve environment variables like /chat/completions.
  • Fixed callback environment variable casing.
  • Fixed login page JSON parsing error.
  • Fixed new user route user_id collision handling.
  • Renamed extra_headers to additional_headers.
  • Fixed Vertex multimodal embeddings to support text + base64 image combinations.
  • Added output_cost_per_image_token for gemini‑2.5‑flash‑image models.
  • Fixed MCP list tool to work without a database.
  • Fixed CI/CD mypy, check_code_and_doc_quality, and guardrails_testing failures.
  • Fixed CI/CD litellm_mapped_tests_llms, litellm_mapped_tests_core, and caching unit tests.
  • Fixed nested spans handling for Arize.

🔧 Affected Symbols

get_model_from_requestget_provider_chat_configAzureAnthropicConfigextra_headersadditional_headerslitellm_paramsmcp_list_toolarize_nested_spanspricing.add_output_cost_per_image_tokenAzureSentinelLoggerLinkUpSearchProviderGuardrailsVertexMultimodalEmbeddingDockerImagePrometheusMetadataExtractor