LiteLLM v1.88.0 ships Gemini 3.5 Flash support and MCP fixes
LiteLLM `v1.88.0` adds day-0 support for Gemini 3.5 Flash, Gemini managed agents, and a cost map for `gemini-3.1-flash-lite`, alongside a range of bug fixes for Bedrock, DeepSeek, MCP JWT auth, and Redis spend counting.
Score breakdown
Teams running LiteLLM in multi-pod deployments should note the Redis spend counter fix, while those integrating Gemini or MCP tooling benefit from day-0 model support and corrected JWT auth.
- 01Day-0 support added for Gemini 3.5 Flash and Gemini managed agents.
- 02Cost map entry added for `gemini-3.1-flash-lite`.
- 03`team_id` and `team_alias` are now propagated to all child OpenTelemetry spans.
LiteLLM `v1.88.0` lands several new features centered on Google's latest models and improved observability. The release adds day-0 support for Gemini 3.5 Flash and Gemini managed agents, and includes a cost map for `gemini-3.1-flash-lite`. The Interactions API endpoint has been added to the UI playground with SSE streaming, and the release migrates to the Google Interactions API steps schema from May 2026. OpenTelemetry tracing is improved by propagating `team_id` and `team_alias` to all child OTEL spans, while Prometheus metrics now include `user_email` and `user_alias` on user budget metrics.
On the bug-fix side, a significant Redis spend counter issue is resolved — a `SET NX` seed strategy now prevents cross-pod double-seeding.
On the bug-fix side, a significant Redis spend counter issue is resolved — a `SET NX` seed strategy now prevents cross-pod double-seeding. MCP JWT authentication is fixed for both `tools/list` and REST `tools/call` server resolution. DeepSeek integration is updated to use the native `/anthropic/v1/messages` endpoint with tool sanitization. Bedrock fixes include correcting `embedding_types` to serialize as a JSON array rather than a string, sanitizing batch metadata to prevent Pydantic `ValidationError`, and switching Bedrock/SageMaker response streaming to lazy loading. The proxy also gains a fix to gate team `allowed_passthrough_routes` to proxy admins only, and SSE pass-through for Google-native `streamGenerateContent` is restored. All Docker images are signed with cosign, verifiable via a pinned commit hash (`0112e53`) or the release tag.
Key facts
- 01Day-0 support added for Gemini 3.5 Flash and Gemini managed agents.
- 02Cost map entry added for `gemini-3.1-flash-lite`.
- 03`team_id` and `team_alias` are now propagated to all child OpenTelemetry spans.
- 04`user_email` and `user_alias` added to Prometheus user budget metrics.
- 05Redis spend counter fixed to use `SET NX` seeding, preventing cross-pod double-seeding.
- 06MCP JWT authentication fixed for `tools/list` and REST `tools/call` server resolution.
- 07Bedrock `embedding_types` corrected to serialize as a JSON array instead of a string.
- 08All Docker images are signed with cosign, verifiable against pinned commit `0112e53`.
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 8, 2026 · 15:36 UTC. How this works →