LiteLLM 1.108.2 (LTS)
LiteLLM 1.108.2 (LTS) addresses 11 bug fixes and 1 update across providers, fixing auth header issues, Bedrock API handling, MCP route failures, and Postgres restart dropouts.
Why this signal matters
LiteLLM release 1.108.2 is an LTS maintenance patch resolving 11 fixes and 1 change. Key updates include: fixing OpenAI deployments with empty API keys defaulting to OPENAI ADMIN KEY bearer tokens; restricting AWS Bedrock Converse service tiers to reserved, auto, and standard while returning 400 for unsupported tiers; fixing Bedrock Invoke parameter drops for the per-turn-control-2026-07-01 beta on Claude models; correcting Vertex AI Chirp TTS region routing; resolving 400 errors for response format: json object on OpenAI/Azure responses endpoints; addressing Claude base64 PDF and Bedrock JSON schema thinking drops; correcting 500 errors on /v1/responses/compact during-call guardrails; handling transcription error leakage; updating hosted MCP tool routing behavior to return 400 or drop invalid parameters via drop params; and fixing routing tag mismatches, prompt caching pinning, and Postgres restart-related spend dropouts.
Actionable summary
LiteLLM LTS patch 1.108.2 introduces fixes for Bedrock Converse tier handling, Claude beta parameters, OpenAI Responses JSON format errors, and MCP tooling fallback handling (returning clear 400s or honoring drop params).
- Agent usefulness
- 57/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade to image tag ghcr.io/berriai/litellm:1.108.2 or pull ghcr.io/berriai/litellm:lts.
- Verify the release artifacts using cosign if container integrity verification is required by policy.
- Review MCP tool configurations if using translated routes that rely on drop params or return 400 responses.