v1.94.0
📦 litellmView on GitHub →
✨ 12 features🐛 21 fixes🔧 26 symbols
Summary
This release introduces enhanced Docker image signature verification using cosign and includes numerous features and bug fixes across the UI, router, and various provider integrations. Key updates involve improved routing capabilities, better error handling, and UI refinements.
Migration Steps
- Verify Docker images using cosign with the pinned commit hash (recommended) or release tag.
- Update the `pillow` dependency to `12.3.0` to resolve CVEs.
✨ New Features
- Working Test Connection for the complexity auto router.
- Adopt openapi-react-query and convert useCustomers.
- Random-pick multi-model complexity tiers.
- Add Router(plugins=[...]) routing-plugin pipeline.
- Soft-floor adaptive mode for complexity router.
- Track cost for unmanaged Bedrock batches, generalize the flag.
- Mint gateway-bound envelope at the token endpoint for dcr_bridge oauth_delegate.
- Send source.type for Used By attribution.
- Interactive SSO sign-in for dcr_bridge oauth_delegate DCR clients.
- Client-held refresh envelope for the dcr_bridge oauth_delegate flow.
- Rebuild the Virtual Keys table on the shared DataTable.
- Rebuild the Teams table on the shared DataTable.
🐛 Bug Fixes
- Use StandardLoggingGuardrailInformation in logging hook.
- Stop unrecognized model namespaces slipping through provider wildcard keys.
- Sanitize scan result before recording it for logging.
- Filter embedding models in complexity tab dropdowns, require all tiers, inline validation.
- Translate raw adaptive thinking for pre-4.6 models on chat completions and Bedrock Converse.
- Add supports_reasoning: false to Gemini image generation models.
- Walk custom_tool_call_output items in _content_utils.
- Show and allow editing team model aliases after team creation.
- Surface rejected delegate-auth upstream tokens as connect-time 401.
- Track unauthenticated pass-through requests in spend logs.
- Continue MCP gateway tool turns from the final response and surface failures.
- Forward aws credential kwargs into litellm_params so the responses bridge keeps WIF auth.
- Respect litellm_key_header_name in BYOK credential save and workflow runs fetches.
- Redact async complete streaming response for custom callbacks.
- Clamp max_output_tokens below API minimum.
- Read v3 rate limiter remaining values for per-key model gauges.
- Drop w-full from page-content wrappers to remove 32px horizontal overflow.
- Correct Responses API tool_choice shape and propagate alias litellm_params.
- Render the sidebar scrollbar with shadcn ScrollArea.
- Relay upstream OAuth token and DCR rejections instead of a generic 500.
- Persist key_type so the UI shows correct key scope instead of "All Proxy Models".