BerriAI/litellm v1.107.0-rc.1 on GitHub

Verify Docker Image Signature

All LiteLLM Docker images are signed with cosign. Every release is signed with the same key introduced in commit 0112e53.

Verify using the pinned commit hash (recommended):

A commit hash is cryptographically immutable, so this is the strongest way to ensure you are using the original signing key:

cosign verify \
  --key https://raw.githubusercontent.com/BerriAI/litellm/0112e53046018d726492c814b3644b7d376029d0/cosign.pub \
  ghcr.io/berriai/litellm:v1.107.0-rc.1

Verify using the release tag (convenience):

Tags are protected in this repository and resolve to the same key. This option is easier to read but relies on tag protection rules:

cosign verify \
  --key https://raw.githubusercontent.com/BerriAI/litellm/v1.107.0-rc.1/cosign.pub \
  ghcr.io/berriai/litellm:v1.107.0-rc.1

Expected output:

The following checks were performed on each of these signatures:
  - The cosign claims were validated
  - The signatures were verified against the specified public key

What's Changed

fix(ui): place Home Create New Virtual Key button like the Virtual Keys page by @krrish-berri-2 in #46778
fix(mcp): scope gateway OAuth consent to x-mcp-servers and add explicit scope setting by @devin-ai-integration[bot] in #44957
fix(mcp): request offline access from Google upstream OAuth providers by @devin-ai-integration[bot] in #41197
fix(anthropic): drop unknown legacy thinking keys under drop_params when rebuilding adaptive thinking by @mateo-berri in #46741
feat(mcp): enable verified stateless protocol capabilities by @joshua-berri in #46264
fix(ci): restore loadscope on the mcp-interactions shard so its MCP subscription cases do not race across workers by @mateo-berri in #46783
feat(proxy): add agent_access_default_deny for least-privilege agent access by @devin-ai-integration[bot] in #43979
fix(rust-bridge): honor the configured request_timeout on native Messages by @yujonglee-berri in #46749
ci: run mcp interactions and redis cluster iam integration tests only in circleci by @yassin-berriai in #46780
test(integration): hand every harness server a port from one per-worker block by @yuneng-berri in #46761
fix(terraform): stop the perpetual plan diff on litellm_mcp_server by @yassin-berriai in #46781
test(e2e): perf users start staggered with jitter, tool-call streams get a first token, small_bodies_5rps sends from the mixed case's callers by @kerry-berri in #46776
fix(ollama): prevent reused response ids from dropping spend logs by @devin-ai-integration[bot] in #39227
fix(vertex-ai): end free period for deepseek-v4.1-flash-maas by @berriai-litellm-provider-info-sync[bot] in #46793
feat(cost-map): add baseten Muse Spark Open and Nemotron 3.5 Super VL by @berriai-litellm-provider-info-sync[bot] in #46794
fix(proxy): limit gzip to dashboard analytics and offload compression by @tin-berri in #46712
chore(pricing): add github_copilot deprecation dates for claude-opus-5 and gpt-5.6-terra by @berriai-litellm-provider-info-sync[bot] in #46792
ci(cost map): classify token_count_framing and token_count_fallback_model in the schema generator by @mateo-berri in #46784
fix(terraform): accept 2xx from the proxy in mcp server, model, key and org requests by @yassin-berriai in #46550
fix(proxy): cooldown errors carry 429, Sentry starts at boot without request bodies, tracing failures log at ERROR, exporter counters exported by @ryan-crabbe-berri in #46344
feat(wandb): add NVIDIA Nemotron 3.5 Super VL to the cost map by @berriai-litellm-provider-info-sync[bot] in #46795
fix(bedrock): forward Nova Sonic realtime tool calls with their arguments by @mrinal-berri in #46214
test(ui): send Connection: close on every integration browser request by @mateo-berri in #46805
refactor(rust-bridge): move the route host modules under host/routes by @yujonglee-berri in #46786
refactor(rust-bridge): let each route host pick the body params by @yujonglee-berri in #46787
fix(responses): preserve explicit websocket API keys (internal copy of #40853) by @mateo-berri in #46724
test(integration): build OpenAI and Anthropic SDK models before tests start threads by @yucheng-berri in #46797
fix(guardrails): honor streaming settings on the Noma v2 guardrail by @yucheng-berri in #46606
fix(count_tokens): count Anthropic-format requests for OpenAI the way /v1/messages inference converts them by @mateo-berri in #46759
fix(sso): share SAML sign-in state over Redis without enable_redis_auth_cache by @yuneng-berri in #46807
fix(responses): write one spend-log row per native WebSocket turn by @kerry-berri in #46768
fix(crusoe): update base URL to current endpoint by @acheamponge in #33121
test(e2e): attach perf case metrics and limits as JUnit properties by @narmstrong-berri in #46774
fix(terraform)!: send additional_drop_params to the proxy instead of deleting the keys it names by @mateo-berri in #46782
fix(count_tokens): count Anthropic-format requests for Gemini the way /v1/messages inference converts them by @mateo-berri in #46767
fix(proxy): reset end-user spend in committed pages so the budget cascade survives a large LiteLLM_EndUserTable by @kerry-berri in #46502
test(e2e): perf datastore limits count the Postgres statements and Redis commands a request waits on, background work reported on its own by @kerry-berri in #46798
fix(cost): bill Responses API image generation and usage reported web search as tool usage cost by @devin-ai-integration[bot] in #41877
fix(count_tokens): return a 400 naming Anthropic blocks the OpenAI and Gemini count can't convert by @mateo-berri in #46772
fix(vertex-ai): drop cached flex price from gemini 3 preview rows by @berriai-litellm-provider-info-sync[bot] in #46826
fix(proxy): only release the legacy limiter's slot from an admitted Responses WebSocket by @devin-ai-integration[bot] in #46822
test(e2e): mixed_size_requests_2rps perf case (coding-agent bodies 5k to 1m tokens with tools, JWT and key callers, two guardrails) by @kerry-berri in #46717
fix(ui): validate the Add Model form when Test Connect is clicked by @yuneng-berri in #46834
fix(bedrock-passthrough)!: return AWS-shaped errors so AWS SDKs read the code and message by @yucheng-berri in #46522
test(integration): wait for the gen-AI span before reading the excluded-services operator trace by @yucheng-berri in #46846
feat(soniox): decode proxy form fields for context, translation and flags by @dan2k3k4 in #40605
fix(router): let an exact base_model widen a deployment's context window by @setarm01 in #46197
fix(bedrock): grant sts:GetWebIdentityToken and sts:TagGetWebIdentityToken in web identity session policy for claude_platform by @Bochenski in #46670
fix(proxy)!: count endpoints return errors in their source API shape instead of silent local counts by @mateo-berri in #46815
fix(responses): preserve provider request IDs in streaming failure logs by @moyai-berriai[bot] in #45453
fix(guardrails): record every content filter match in logging_only and offer the mode in the UI by @yucheng-berri in #46673
fix(anthropic): keep trailing user messages with None or empty list content by @devin-ai-integration[bot] in #41397
fix: preserve non-null anyOf branches when stripping null types for Vertex by @debanshd in #41683
perf(ui): page the member tables and look up budget rows once per render by @Bytechoreographer in #26667
fix(proxy): track and drain management audit log writes on shutdown by @akram1089 in #43594
feat(gemini): allow disabling thought summaries while keeping thinking on by @Aryan-Pardeshi in #36460
fix(router)!: keep RPM, region, and param checks when token counting fails by @ryuto-takemoto in #44544
build(deps): bump claude-agent-sdk to 0.2.165 to unblock osv-scan (rc/1.107.0) by @yuneng-berri in #46976
fix(mcp): stop proxy-started legacy listeners from consuming server rpm (rc/1.107.0) by @yuneng-berri in #46982
fix(cost): bill bridged /v1/messages Responses replies like native /v1/responses for reasoning tokens and service tier (backport #46975 to rc/1.107.0) by @yuneng-berri in #47009
fix(proxy): backport the 1.107.0 rc triage fixes to rc/1.107.0 by @yuneng-berri in #47043
fix!: backport the second batch of rc/1.107.0 CI triage fixes from main by @yuneng-berri in #47090
feat(mcp): backport the ID-JAG stack to rc/1.107.0 (#46748, #46747, #46745, #47004) by @yassin-berriai in #47058
fix(mcp)!: backport the ID-JAG signed resource claim cap to rc/1.107.0 (#47094) by @yassin-berriai in #47138
fix(mcp)!: backport the OAuth Delegate fields and gateway-owned DCR bridge client to rc/1.107.0 (#46941, #46940) by @yassin-berriai in #47153
test: backport the SDK logging drain to rc/1.107.0 (#46924) by @yuneng-berri in #47143
fix(vector-stores): backport #47021 to rc/1.107.0 by @yuneng-berri in #47159
fix: backport the third batch of rc/1.107.0 CI triage fixes from main by @yuneng-berri in #47174
fix(bedrock): backport #46307 and #46524 to rc/1.107.0 by @yucheng-berri in #47157
feat(count_tokens)!: backport the count_tokens stack to rc/1.107.0 (#46839, #46843, #46583, #47057, #47064, #47163, #46777) by @yuneng-berri in #47209
chore: bump litellm-enterprise 0.1.77 -> 0.1.78, litellm-proxy-extras 0.4.109 -> 0.4.110 on rc/1.107.0 by @yuneng-berri in #47230
fix(ui): backport #47025 to rc/1.107.0 by @yuneng-berri in #47224
chore: rebuild Admin UI bundle for rc/1.107.0 by @yuneng-berri in #47232
fix(guardrails)!: backport #45874 to rc/1.107.0 by @yucheng-berri in #47234

New Contributors

@setarm01 made their first contribution in #46197
@debanshd made their first contribution in #41683
@akram1089 made their first contribution in #43594
@Aryan-Pardeshi made their first contribution in #36460
@ryuto-takemoto made their first contribution in #44544

Full Changelog: v1.107.0-dev.3...v1.107.0-rc.1