github BerriAI/litellm v1.111.0-rc.3

pre-release58 minutes ago

Verify this release

cosign verify \
  --certificate-identity https://github.com/BerriAI/project-releaser/.github/workflows/release.yml@refs/heads/main \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com \
  ghcr.io/berriai/litellm:v1.111.0-rc.3

Same command for litellm-database, litellm-non_root and the docker.io/litellm/* images

What's new in 1.111.0-rc.3

15 PRs. 11 fixed. No action needed to upgrade

Changes

  • Anthropic: Fixed temperature=1 returning 400 on Claude models that removed sampling params on Anthropic, Vertex AI and Azure AI, broken since v1.110.0-rc.2 (#52341) Also in 1.110.2
  • Responses API: Fixed streams retrying after output reached the client, which sent duplicate events and billed extra calls (#52461) Also in 1.110.2
  • Responses API: Fixed streams that fail after reasoning output restarting on a fallback group and sending a second response (#52616) Also in 1.110.2
  • Caching: Fixed GPT-6.2 system prompts missing the prompt cache on /v1/chat/completions and /v1/messages unless store: false (#52906) Also in 1.106.12, 1.108.13, 1.109.3, 1.110.2
  • Cost tracking and pricing: Fixed forwarded Anthropic credentials on /v1/messages ignoring the deployment's custom prices, broken since v1.109.0 (#52313) Also in 1.110.2
  • Keys, teams and auth: Fixed 503s on a burst of first requests for new customer org aliases, which left their spend untracked (#52080)
  • Keys, teams and auth: Fixed JWT key mappings without jwt_issuer taking 5 minutes to apply when litellm_jwtauth.issuers is set, broken since v1.103.0 (#52350) Also in 1.108.13, 1.109.3, 1.110.2
  • Keys, teams and auth: Fixed /key/info letting a key with no user read other userless keys by hash or alias (#52838, #52925) Also in 1.106.12, 1.108.13, 1.109.3, 1.110.2
  • Admin UI: Fixed SSO login redirecting to an http:// URL when the proxy runs behind TLS termination, broken since v1.108.0 (#52213)
  • Deployment: Fixed each proxy worker opening six extra Postgres connections at boot (#52378)
  • Performance: Fixed failed calls holding large request bodies in memory until a full GC, growing workers to several GiB (#52600)

Backport #52462 · Backport #52470 · Backport #52691 · Backport #52752 · Backport #52764 · Backport #52914 · Backport #52939 · Backport #53024 · Full changelog

Don't miss a new litellm release

NewReleases is sending notifications on new releases.