v1.84.0
Breaking Changes📦 litellmView on GitHub →
⚠ 1 breaking✨ 11 features🐛 23 fixes🔧 10 symbols
Summary
This release introduces new features like support for gpt-image-2 and AIHubMix provider, alongside numerous bug fixes across caching, logging, and provider integrations. It also contains a breaking change: the removal of the /ui/chat page.
⚠️ Breaking Changes
- The /ui/chat page has been removed. Users relying on this specific UI endpoint must update their access methods.
Migration Steps
- If you relied on the /ui/chat page, you must update your access methods as it has been removed.
✨ New Features
- Added support for gpt-image-2 model.
- Added AIHubMix as a new OpenAI-compatible provider.
- Added support for Azure Entra discovery endpoint.
- Added durable agent workflow run tracking via /v1/workflows/runs endpoint.
- Added support for Bedrock retrievalConfiguration passthrough in vector stores.
- Added opt-in short-ID tool prefix for MCP to keep tool names under the 60-character limit.
- Added LiteLLM headers on Google native generateContent routes.
- Added optional TCP SO_KEEPALIVE support to aiohttp's TCPConnector.
- Added --timeout_worker_healthcheck flag for uvicorn worker triage in the proxy.
- Added cleanup job for expired LiteLLM dashboard session keys in the proxy.
- Added retry settings for generic API logger.
🐛 Bug Fixes
- Fixed GCP IAM token caching in Redis to prevent async event loop blocking.
- Fixed NOMA v2 deepcopy crashing in build scan payload.
- Fixed UI to use stored-credentials endpoint for tools fetch on MCP edit page.
- Fixed CI to support CircleCI rerun failed tests for local_testing jobs.
- Fixed GPT-5.5 Pro Pricing.
- Fixed UI to move 'Store Prompts in Spend Logs' toggle to Admin Settings.
- Fixed preserving prompt_tokens_details through embedding cache round-trip.
- Fixed backfilling streaming hidden response cost in logging.
- Fixed Vertex AI to reuse Anthropic messages config instances.
- Fixed Vertex AI to preserve items on array branches in anyOf with null and de-flaked a test.
- Fixed replacing deprecated Bedrock Claude 3.7 Sonnet model ID.
- Fixed caching LiteLLM_Config param reads in DualCache and batch.
- Fixed unifying cost calculation in success_handler dict and typed branches.
- Fixed scope of ERROR log assertion to LiteLLM logger in test_model_alias_map.
- Fixed translating OpenAI file content on tool-result path for Bedrock and Anthropic.
- Fixed test for Vertex batches by setting is_redirect=False on mocked retrieve response.
- Redacted spend logs error message.
- Fixed proxy self-healing for Prisma read paths and hardened reconnect state machine.
- Dropped sensitive locals from re-raised error messages.
- Redacted credentials in vector store list/info/update responses and gated update by per-store access.
- Fixed proxy to inherit caller identity in passthrough batch managed-object.
- Fixed team member null budget fallback.
- Fixed proxy/auth to tighten guardrail m...