v1.83.14-stable
📦 litellmView on GitHub →
✨ 9 features🐛 25 fixes🔧 13 symbols
Summary
This release focuses heavily on security by introducing Docker image signing verification via cosign and includes numerous fixes across various providers (Anthropic, Azure, Gemini, Bedrock) and proxy features, alongside new model support like GPT-5.5.
Migration Steps
- Users running Docker containers should verify image signatures using the provided cosign commands to ensure image integrity.
✨ New Features
- Added Docker image signing using cosign for enhanced security verification.
- Added GLM-5 and Minimax M2.5 models to Bedrock with regional aliases.
- Added versioned GPT-5.4 mini/nano snapshots.
- Added Day-0 support for GPT-5.5 and GPT-5.5 Pro.
- Added LLM-as-a-Judge guardrail.
- Added use_chat_completions_api flag for openai/ models with custom api_base in the Responses API.
- Added route_all_chat_openai_to_responses global flag for OpenAI.
- Added Send Invitation Email Toggle in the UI for Users.
- Added GA cost map, blog, and test for gemini-embedding-2 pricing.
🐛 Bug Fixes
- Fixed preservation of tool_use input args in Anthropic adapter streaming.
- Fixed preservation of role='assistant' in Azure streaming when include_usage is set.
- Fixed mapping of Zhipu GLM non-standard finish_reason values.
- Applied GPT-5 temperature validation in the Responses API.
- Fixed sorting of assistant content blocks in Bedrock so text precedes toolUse.
- Fixed filtering of parameters from Gemini embedding requests.
- Fixed reading web search cost from model_info instead of hardcoding for Gemini.
- Fixed inclusion of DOCUMENT modality tokens in Gemini cost calculation.
- Fixed forwarding of dimensions parameter in Vertex AI multimodal embedding requests.
- Migrated 38 models from legacy max_tokens to max_input_tokens/max_output_tokens in model prices.
- Updated Bedrock Claude Sonnet/Opus 4.6 above 200k token pricing and set Sonnet 4.6 max_input_tokens to 1M.
- Restored BYOK key injection for vector store endpoints with team-scoped deployments in the router.
- Split MCP routes into inference vs management to unblock Admin UI on DISABLE_LLM_API_ENDPOINTS nodes.
- Auto-added SSO team members to org on move for team endpoints (proxy admin only).
- Fixed respecting object-level permissions for managed vector store endpoints in the proxy.
- Normalized bridged object field in responses.
- Preserved anthropic_messages call type for /v1/messages logging in the proxy.
- Stripped custom_tool_call namespace for all providers in responses.
- Stripped Gemini thought suffix from streaming tool_use id in Anthropic adapter.
- Centralized common_checks to close authorization bypass.
- Hardened OAuth authorize/token endpoints (BYOK + discoverable) in MCP.
- Fixed logging of :embedContent and :batchEmbedContents responses for Vertex AI passthrough.
- Applied team TPM/RPM + attribution for admins using x-litellm-team-id in JWT authentication.
- Fixed Guardrail parameter handling in list and submission endpoints.
- Fixed proxy single-team DB fallback when JWT has...