v1.89.4 - Vertex Batch Uploads & CVE Patches
We're investigating a potential throughput regression in recent releases. It appears limited to certain deployment configurations rather than affecting all deployments, so we expect it to impact only a small subset of users; correctness and error rates are not affected.
Update (June 26, 2026): We've identified potential causes and will need more time to test before we can confirm a fix. We'll update this note with the affected versions and the fix as soon as we have them.
For throughput-sensitive workloads, we recommend validating performance in a staging environment before rolling out an upgrade.
Deploy this version​
- Docker
- Pip
docker run \
-e STORE_MODEL_IN_DB=True \
-p 4000:4000 \
docker.litellm.ai/berriai/litellm:1.89.4
pip install litellm==1.89.4
v1.89.4 is a patch release on top of v1.89.3. It streams OpenAI→Vertex batch JSONL uploads instead of buffering them in memory, backports cost-tracking recovery for interrupted Anthropic streams, adds a no-mcp-servers sentinel that scopes a key to zero MCP servers, and clears the remaining OSV-flagged CVEs with OpenSSL and dependency bumps. The bundled litellm-enterprise package is bumped to 0.1.42.post2.
What's Changed​
- fix(passthrough): recover output tokens for interrupted anthropic streams - PR #30787
- fix(proxy): record partial spend on the failure row for interrupted streams - PR #30788
- feat(mcp): scope a key to zero MCP servers with no-mcp-servers sentinel - PR #31029
- fix(passthrough,streaming): recover cost on interrupted and agentic Anthropic streams - PR #31035
- fix(vertex/files): stream OpenAI->Vertex batch JSONL uploads - PR #31036
- fix(deps): bump osv-flagged dependencies to clear known CVEs - PR #31122
- fix(docker): bump wolfi-base digest to patch openssl CVE-2026-34182 - PR #31133
Full Changelog​
https://github.com/BerriAI/litellm/compare/v1.89.3...v1.89.4