A Guth Labs publication ↗
Back to News

LiteLLM updates its LTS release with caching and request-handling fixes

Cream conduit junction with transparent green channels and a gold control valve.
AI-generated editorial illustration

LiteLLM released version 1.108.12 of its long-term-support branch on November 19, fixing caching, routing and request-handling problems across several AI providers.

The official release notes describe a fix for GPT-6.2 system-prompt caching in certain OpenAI-compatible requests. Another change stops upstream OpenAI streaming work from continuing after a client closes its stream through the Messages API.

The patch also addresses Azure model-router failures with Responses requests and a Vertex AI batch-processing issue involving structured-output schemas. Messages API compatibility and tool-policy enforcement receive additional corrections.

Maintainers describe the release as eight pull requests covering six fixes and say no extra action is required to upgrade. Those statements describe the project’s patch guidance, not independent testing of every deployment or a guarantee that an application’s configuration will remain unaffected.

For teams using LiteLLM to connect applications with different model providers, the update targets integration behavior rather than model capability. Correct caching, cancellation and policy enforcement matter because they shape how requests consume resources and how consistently gateway rules apply across providers.

Sources and citations

Revision history

Published .