BerriAI/litellm v1.107.0-dev.2 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-dev.2

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-dev.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.107.0-dev.2

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

feat(claude-code): emit installationPreference in marketplace.json entries by @devin-ai-integration[bot] in #41430
fix(proxy): scope /global/activity/cache_hits analytics to the caller's own and permitted-team logs by @ojensen-berri in #46210
fix(anthropic): cache tool result screenshots on the /v1/messages to Responses bridge by @ryan-crabbe-berri in #46324
test(integration): cover fine_tuning endpoint contracts by @devin-ai-integration[bot] in #44697
fix(databricks): send a single plain text block as a string for non-Claude models by @mrinal-berri in #45686
feat(ui): show decision models as decisions and list the supported ones in the AI Hub by @devin-ai-integration[bot] in #45965
feat(decisions)!: send images to Cloudflare Clef through its images field by @devin-ai-integration[bot] in #45734
feat(anthropic): generate auto mode safeguard_results with a classifier model on the /v1/messages chat bridge by @devin-ai-integration[bot] in #45600
feat(mcp): separate confidential client for ID-JAG leg 2 by @joshua-berri in #46211
fix(http_handler): only replay POST and PATCH when the request never reached the upstream by @mateo-berri in #46336
feat(decisions): add Laya and Bespoke providers to the Add Model form by @devin-ai-integration[bot] in #45260
fix(ui): widen the LiteAdmin home model selector by @moe-berri in #46446
fix(cache): keep cache hits free of unset fields, and follow intentional behavior changes in CircleCI suites by @mateo-berri in #46359
fix(anthropic): check Vertex forced-tool support on the real model, skip non-string user ids, drop speed where unsupported by @ryan-crabbe-berri in #46337
fix(sso)!: reject empty identities before session creation by @yucheng-berri in #44098
chore(codeowners): replace kerrylu-berri with kerry-berri by @kerry-berri in #46452
fix(cost-map): registry audit 2026-10-14, remove azure_ai/claude-fable-5-5 by @devin-ai-integration[bot] in #46428
fix(proxy): build configured logging callbacks at boot so a broken callback fails startup instead of the first request by @ryan-crabbe-berri in #46378
feat(dashscope): add Wan 3.0, Wan 2.7 and HappyHorse video generation via /v1/videos by @Bytechoreographer in #43706
fix(anthropic): forward service_tier on chat completions to Anthropic and Azure by @ryan-crabbe-berri in #46461
fix(ui): label Decisions playground issues by question key and fix the Response header row by @mateo-berri in #46145
fix(http): restore the 15s aiohttp keep-alive default in the Python and Rust clients by @mateo-berri in #46355
fix(vertex_ai): pass the per-request timeout to Gemini streaming calls by @kerry-berri in #46349
feat(proxy): add deployment warning endpoints with instance-wide ignore and snooze by @yuneng-berri in #46347
fix(proxy): check previous_response_id ownership on /v1/responses/compact by @mateo-berri in #46408
refactor(router): name the pre-content stream retry helpers for both stream paths and let the retry loop take its attempt function by @kerry-berri in #46474
fix(anthropic): stream json_schema answers as content on Bedrock Mantle and Claude Platform, and join split answers by @mateo-berri in #46331
fix(caching): collect the Redis circuit-breaker state gauge as one series per state across worker processes by @kerry-berri in #46357
fix(gemini): add the changelog shutdown date to gemini-2.0-flash-exp-image-generation by @berriai-litellm-provider-info-sync[bot] in #46489
test(integration): remove six setup causes of CircleCI integration reds by @yuneng-berri in #46490
fix(s3_v2): bound each upload attempt, add jitter, retry 429 and transport errors in the call by @ryan-crabbe-berri in #46340
fix(streaming): add CustomStreamWrapper.close() and release the connection when a sync stream ends by @kerry-berri in #46354
feat(proxy): let proxy admins open soft_budget to team admins by @yuneng-berri in #46485
fix(proxy): accept every known logging callback on /health/services by @yassin-berriai in #46297
test(proxy): cover the OTel V2 otel callback on /health/services by @yassin-berriai in #46394
fix(vllm): write spend log rows for /vllm passthrough calls to router models by @devin-ai-integration[bot] in #45733
fix(config): treat blank env values as unset so the proxy neither crash-loops nor silently drops a guardrail by @ryan-crabbe-berri in #46341
fix(proxy): evict the cached user on every proxy for every /user/update and /user/bulk_update by @yuneng-berri in #46500
feat(proxy): let proxy admins open team member defaults to team admins by @yuneng-berri in #46353
fix(anthropic): forward service_tier on /v1/messages for anthropic, azure_ai and bedrock_mantle by @kerry-berri in #46370
fix(router): read the first stream chunk before returning so a failure before any token runs through the retry and fallback budget by @kerry-berri in #46493
fix(caching): report Redis circuit-breaker metrics through the Prometheus services logger by @kerry-berri in #46484
fix(anthropic): keep refusal stop_reason and stop_details on the logged response by @devin-ai-integration[bot] in #46104
fix(router): add fallback_on_unsupported_params so param rejections can raise instead of falling back by @kerry-berri in #46350
feat(access-groups): per-server MCP tool permissions on access groups by @yassin-berriai in #46300
feat(helm): read database and Redis host or URL from a Kubernetes Secret by @yassin-berriai in #46410
fix(mcp): leave out tool texts too long to embed instead of failing tool search by @devin-ai-integration[bot] in #43554
perf(scim): batch group member identity lookups into one query per push by @yassin-berriai in #46412
feat(helm): support a separate hostname per component in the componentized deploy by @yassin-berriai in #46360
fix(redis): default every Redis client to a 5s socket timeout when none is set by @ryan-crabbe-berri in #46501
fix(router): report fallbacks skipped for access or budget instead of calling them failed by @yassin-berriai in #46291
fix(proxy): finish prisma engine requests when their caller is cancelled by @yassin-berriai in #46503
fix(proxy-extras): run prisma migrate deploy under the pinned coordinator lock instead of Prisma's session advisory lock by @yassin-berriai in #46368
test(integration): fix three more CircleCI integration reds from test setup by @yuneng-berri in #46510
fix(caching): make Redis Cluster commands wait for a pool slot instead of tripping the circuit breaker by @yassin-berriai in #46488
test(rust): assert native serving via the x-litellm-rust response marker by @yujonglee-berri in #46531
fix(responses): propagate cache_control through responses->chat completion bridge by @TobiMayr in #32511
feat(guardrails): add Airia guardrail integration by @airia-open-source in #40784
feat(terraform): add team_member_model_max_budget to litellm_team by @matthowardcohere in #46274
fix(responses): keep the provider code and message on top-level stream error events by @mateo-berri in #46438
feat(access-groups): grant Claude Code skills through access groups by @yassin-berriai in #46294
test(integration): delete flaky team model TPM concurrency test by @kerry-berri in #46541
fix(vertex-ai): add flex prices to gemini-nano-banana-2.1 by @berriai-litellm-provider-info-sync[bot] in #46512
feat(proxy): let proxy admins open team metadata to team admins, keeping proxy settings with proxy admins by @yuneng-berri in #46506
fix(responses): return the first message item as content for structured output on the chat bridge by @kerry-berri in #46508
fix(cost): split xai spend rows across input and output instead of booking the total as output by @kerry-berri in #46320
test(rust-bridge): pin Messages and Responses callback contracts on the native path by @yujonglee-berri in #46443
test(host-python): move driver tests out of the inline module by @yujonglee-berri in #46325
refactor(host-python): collect host failures in a thiserror Error enum by @yujonglee-berri in #46361
refactor(host-python): run the Python driver as one async body by @yujonglee-berri in #46235
refactor(rust-bridge): generator-style execution protocol with stream step names by @yujonglee-berri in #46282
test(translation): share rich request and response templates by @kerry-berri in #46546
refactor(rust-bridge): return an unstarted Execution from native entrypoints and drive it in Python by @yujonglee-berri in #46299
test(integration): add claude-haiku-5-5 basic translation cases by @devin-ai-integration[bot] in #45176
feat(ui): add the Airia guardrail to the Guardrail Garden with its logo by @airia-open-source in #46562
feat(proxy): return the W3C traceresponse header on traced responses by @adityajoy-1902 in #42079
feat(proxy): general_settings.disable_requester_ip_address_logging keeps caller IP out of spend logs by @yucheng-berri in #45804
test(integration): update stale CircleCI integration expectations by @yuneng-berri in #46570
fix(cli): run group subcommands in the lite shell and default to the logged-in gateway by @mateo-berri in #46511
fix(cost): price AgentCore web search at $7 per 1,000 queries by @litellm-stampbot in #46571
feat(ui): replace deployment warning banners with a health icon, alert banner and panel by @yuneng-berri in #46498
fix(caching): report zero response_cost on a response-cache hit by @Apeksha9389 in #46471
fix(responses bridge): keep defer_loading on function tools sent to the Responses API by @mateo-berri in #46439
fix(router): read the first stream chunk through aiter so MCP auto-run streaming works again by @kerry-berri in #46578
fix(anthropic): keep tool_search_tool_result so the search call replays with its result by @mateo-berri in #46440
test(translation): add gpt-6-astra translation cases by @devin-ai-integration[bot] in #46580
test(translation): recapture bedrock invoke claude-sonnet-4-6 bases after usage shape change by @devin-ai-integration[bot] in #46579
test(translation): add claude-fable-5-1 translation cases by @devin-ai-integration[bot] in #46581
fix(bedrock)!: replay redacted_thinking on Converse and keep its data as Claude's own string by @ryan-crabbe-berri in #46532
test(translation): share rich fixtures across Claude, GPT and Gemini providers by @kerry-berri in #46585
fix(bedrock): use base_model for Claude capability checks on application inference profile ARNs by @ryan-crabbe-berri in #46569
feat(cost-map): add scx-ai DeepSeek-V4.1-flash, GLM-5.3 and Kimi-K3 by @devin-ai-integration[bot] in #46496
fix(cost): price typesafe/jev-1.14.0 by @hyamie in #46594
fix(responses): translate OpenAI tool_search to Anthropic tool search for Claude models by @devin-ai-integration[bot] in #45882
test(integration): follow #46101 in the breakpoint cache mark census expectation by @mateo-berri in #46509
fix(router): keep the prompt caching affinity pin alive as long as the prefix's cache_control ttl by @devin-ai-integration[bot] in #46084
fix(anthropic): emit one tool_use block per call when a /v1/messages stream chunk carries several by @narmstrong-berri in #46492
refactor: clean up fresh tech debt in per-user Copilot auth, Ternary FOCUS export and telemetry by @devin-ai-integration[bot] in #45769
fix(responses bridge): round-trip tool_search_call and tool_search_output items between turns by @mateo-berri in #46444
fix(anthropic)!: translate the OpenAI tool_search tool on /v1/chat/completions for Claude models by @mateo-berri in #46389
fix(xai): name a completed xAI batch's results file and stop looking it up on xAI's files API by @yucheng-berri in #45559
chore(prices): add xAI grok-voice-3.0, grok-voice-2.0 and grok-voice-latest realtime rows by @berriai-litellm-provider-info-sync[bot] in #46602
fix(responses): map Anthropic tool search to OpenAI server tool search by @mateo-berri in #46441
fix(proxy): store list-valued audit log changes as JSON text by @yuneng-berri in #46555

New Contributors

@airia-open-source made their first contribution in #40784
@matthowardcohere made their first contribution in #46274
@adityajoy-1902 made their first contribution in #42079
@Apeksha9389 made their first contribution in #46471
@hyamie made their first contribution in #46594
@narmstrong-berri made their first contribution in #46492

Full Changelog: v1.107.0-dev.1...v1.107.0-dev.2