v1.86.3 - Gemini 3.5 Flash Day-0 & Pending Line Backports

Deploy this version

Docker

docker run \
-e STORE_MODEL_IN_DB=True \
-p 4000:4000 \
docker.litellm.ai/berriai/litellm:1.86.3

Pip

pip install litellm==1.86.3

v1.86.3 is a patch release on top of v1.86.2. It closes the gap with the 1.84 and 1.85 lines: day-0 Gemini 3.5 Flash on Vertex AI and Google AI Studio with its paired Vertex tool-call fix, Redis spend-counter seeding, and the observability, budget, and flag-leak fixes.

What's Changed

• feat: day-0 support for Gemini 3.5 Flash on Vertex AI and Google AI Studio - PR #28268
• fix(vertex): omit the function_call id on Gemini 3.5+ tool turns (pairs with #28268) - PR #28324
• fix(spend): seed the Redis spend counter with SET NX so concurrent pods no longer double-seed - PR #27854
• fix(logging): stop duplicate Claude Code traces, plus the _build_passthrough_logging_result helper - PR #29311
• fix(proxy): normalize the Bearer prefix in the safe-hash helper - PR #29343
• fix(budget): reset_budget writes only {spend, budget_reset_at} - PR #29358
• fix(proxy): stop the use_chat_completions_api flag from leaking into the provider request body - PR #29447

Full Changelog

https://github.com/BerriAI/litellm/compare/v1.86.2...v1.86.3