Release notes from litellm

v1.105.0-dev.2

2026-10-02T08:33:03Z

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:v1.105.0-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/v1.105.0-dev.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.105.0-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

fix(router): carry per-request routing reads on context variables instead of public method kwargs by @devin-ai-integration[bot] in #43814
fix(bedrock): add beta header for output config in message by @shrey-berri in #43778
refactor: clean up fresh tech debt from 2026-09-29 by @devin-ai-integration[bot] in #43830
test(router): settle the shared logging worker before recording shadow callbacks by @devin-ai-integration[bot] in #43847
chore(model_prices): add Gemini Veo, Mistral and Azure Claude 4.5 deprecation dates by @devin-ai-integration[bot] in #43857
fix(cost_calculator): bill ultrafast prompts above 272k at the ultrafast long-context rates by @devin-ai-integration[bot] in #43764
refactor(rust): centralize Python bridge execution wrappers by @devin-ai-integration[bot] in #43871
chore(cost-map): add openai gpt-image-2.5 batch prices from the pricing page by @berriai-litellm-provider-info-sync[bot] in #43869
chore(cost-map): add fireworks priority prices for ember-1, nemotron and glm 5.3 us rows by @berriai-litellm-provider-info-sync[bot] in #43811
fix(proxy): attribute completed batch cost rows to /batches in daily activity by @devin-ai-integration[bot] in #43870
fix(router): strip encrypted reasoning the pinned deployment cannot decrypt by @devin-ai-integration[bot] in #43781
fix(cost_calculator): stop copying optional_params into response hidden params by @yucheng-berri in #43637
feat(pricing): add vertex_ai gemini-3.8 flash tts rows by @berriai-litellm-provider-info-sync[bot] in #43876
fix(proxy): keep request-body credentials out of stored spend-log requests by @yucheng-berri in #43635
fix(ui): right-align money and count columns across tables by @ryan-crabbe-berri in #37889
feat(agents): enforce authoritative agent permissions by @joshua-berri in #43721
fix(proxy): strip caller credentials from websocket passthrough by @devin-ai-integration[bot] in #43855
fix(ui): render access group MCP and agent selections as wrapping chips by @devin-ai-integration[bot] in #41228
fix(responses): scan and mask top-level instructions with guardrails by @devin-ai-integration[bot] in #43629
feat(traces): add Rust storage foundation by @yujonglee-berri in #43819
feat(ui): agent traces tab on logs with timeline and otel setup guide by @ishaan-berri in #43891
feat(otel v2): excluded_services opt-out for datastore spans on tenant destinations by @devin-ai-integration[bot] in #43278
fix(traces): correct ClickHouse rollup partitioning, dedupe keys, and retention changes by @devin-ai-integration[bot] in #43901
fix(bedrock): keep applicable beta headers by @shrey-berri in #43829
feat(agents): authenticate Entra identities and delegated requests by @joshua-berri in #43722
fix(hosted_vllm): keep reasoning_content on replayed assistant messages by @devin-ai-integration[bot] in #43599
fix(proxy): delete large teams without per-member transaction fan-out by @devin-ai-integration[bot] in #42998
feat(tracing): port OTLP ingestion to current trace foundation by @yujonglee-berri in #43915
feat(proxy): add native ROI calculator for gateway spend vs merged PRs by @devin-ai-integration[bot] in #43669
test(e2e): repair completion, SAIL, and spend-log fixtures by @yuneng-berri in #43902
test(s3_v2): pin async 5xx retry through the production AsyncHTTPHandler by @devin-ai-integration[bot] in #43080
feat(ui): adopt the new LiteLLM logo and monogram by @yuneng-berri in #43913
fix(proxy): relay Azure passthrough body model groups through the router by @devin-ai-integration[bot] in #43896
fix(proxy): register a UI-configured arize callback next to otel under OTel v2 by @devin-ai-integration[bot] in #43906
fix(transcription): honor base_url alias for Groq Whisper and report it as the api base by @devin-ai-integration[bot] in #43917
feat(tracing): store spend in ClickHouse automatically by @yujonglee-berri in #43928
fix(packaging): keep wheel paths under Windows MAX_PATH for Store Python by @ryan-crabbe-berri in #43903
feat(agents): add identity registration and dashboard controls by @joshua-berri in #43723
refactor(proxy): answer every team access check with TeamAccess.allows by @ryan-crabbe-berri in #43364
test(bedrock): restore the AWS env after a failed live call in the auth tests by @devin-ai-integration[bot] in #43921
feat(lens): analyze agent activity with a separate worker by @moe-berri in #43889
fix(router): bill service tiers at catalog rates for custom-priced deployments by @devin-ai-integration[bot] in #43890
test(ci): refresh qualified retired OpenAI fixtures by @yuneng-berri in #43938
fix(anthropic): forward the dangerous-tool-use beta to Azure AI Foundry by @devin-ai-integration[bot] in #43934
test(proxy): scope user_api_key_auth overrides in proxy_server tests by @yuneng-berri in #43952
fix(wandb): set supports_vision true on GLM-5.3-Flash by @berriai-litellm-provider-info-sync[bot] in #43951
chore(cost-map): sync openrouter prices from the models API by @berriai-litellm-provider-info-sync[bot] in #43950
fix(grayswan): send request conversation and tool calls to post-call monitor by @devin-ai-integration[bot] in #43770
fix(azure_storage): name Data Lake objects without base64 padding or slashes by @devin-ai-integration[bot] in #43914
fix(guardrails): treat an unknown straiker api_version as unset instead of skipping the guardrail by @devin-ai-integration[bot] in #43956
chore(deps): bump gitpython and tornado, extend diskcache osv ignore to Nov 1 by @yuneng-berri in #43961
fix(azure_storage): keep the DataLakeServiceClient alive until its TTL elapses by @devin-ai-integration[bot] in #43082
feat(proxy): record in spend logs whether a request used a client-forwarded Anthropic OAuth token by @devin-ai-integration[bot] in #43063
test(ci): repair stale tests and move retired OpenAI text-completion fixtures by @yuneng-berri in #43958
feat(e2e): record each e2e test's steps, starting with ProxyClient by @ryan-crabbe-berri in #42393
feat(providers): add Cortecs as an OpenAI-compatible provider by @devin-ai-integration[bot] in #43872
feat(ui): filter tags by name and description on the Tag Management page by @galovics in #42949
fix(caching): write the response-cache SET to Redis at once instead of on the post-call batch by @devin-ai-integration[bot] in #43973
test(e2e): typed per-test metadata for the e2e suite by @ryan-crabbe-berri in #42044
feat(guardrails): honor litellm_params.timeout in every HTTP guardrail by @devin-ai-integration[bot] in #43134
chore(cost-map): add fireworks inkling priority prices from the prices api by @berriai-litellm-provider-info-sync[bot] in #43949
chore(cost-map): add deprecation date for anthropic claude-sonnet-4-5 by @berriai-litellm-provider-info-sync[bot] in #43898
fix(cost-map): raise baseten DeepSeek-V4.1-Flash max output to 262144 by @berriai-litellm-provider-info-sync[bot] in #43916
fix(proxy): restore pre-config-wins handling of pass-through endpoints by @yuneng-berri in #43962
feat(lens): investigate sampled traces and retain batch results by @moe-berri in #43942
fix(guardrails): scan Responses API input in Azure Prompt Shield by @devin-ai-integration[bot] in #43786
fix(guardrails): scan Responses API input in Azure Text Moderation by @devin-ai-integration[bot] in #43965
refactor: clean up fresh tech debt from 2026-09-30 by @devin-ai-integration[bot] in #43993
test: inject the HIBP client and the MCP loop clock so two backend tests stop flaking by @devin-ai-integration[bot] in #44007
fix(cost-map): reprice fireworks deepseek v4.1 flash to the 2026-10-01 pricing update by @berriai-litellm-provider-info-sync[bot] in #44024
test(anthropic): native /v1/messages reasoning integration tests built on a captured Claude Code request by @devin-ai-integration[bot] in #43361
fix(bedrock): add beta header for thinking display updates by @shrey-berri in #43832
fix(proxy): preserve decision request bodies under token limits by @shrey-berri in #43920
test(proxy): migrate DB and Redis backed proxy tests into tests/integration by @devin-ai-integration[bot] in #43996
feat(lens): track worker spend through virtual keys by @moe-berri in #43989
test(proxy): move auth, hooks, policy_engine and client tests into tests/unit/proxy by @devin-ai-integration[bot] in #43998
fix(ui): give model leaderboard a distinct trophy icon by @moe-berri in #44036
test(ci): repair stale tests and flaky CI infrastructure by @yuneng-berri in #43983
test(proxy): move management_endpoints, management_helpers and guardrails tests into tests/unit/proxy by @devin-ai-integration[bot] in #44003
feat(ui): agent traces open in a side drawer with a chat-style run view by @ishaan-berri in #43972
feat(s3_v2): add s3_partition_granularity option for hourly S3 folders by @devin-ai-integration[bot] in #43748
refactor(lens)!: rename internal engine code and API by @moe-berri in #44034
test(proxy): move utils, agent_endpoints and endpoint tests into tests/unit/proxy by @devin-ai-integration[bot] in #44006
test(proxy): move proxy_server, _experimental and db tests into tests/unit/proxy by @devin-ai-integration[bot] in #44012
test(proxy): move middleware, spend_tracking, pass_through, common_utils and root proxy tests into tests/unit/proxy by @devin-ai-integration[bot] in #44015
test(proxy): delete the legacy proxy test tree and shard tests/unit/proxy by glob by @devin-ai-integration[bot] in #44018
chore(deps): bump pypdf from 6.16.2 to 6.19.0 by @dependabot[bot] in #44033
chore(deps): drop unused pytest-postgresql dev dependency by @yuneng-berri in #44056
docs(proxy): point mcp_server test references at tests/unit/proxy by @yuneng-berri in #44055
feat(ui): show daily token totals on the model leaderboard by @ishaan-berri in #44044
chore(lint): remove the LIT002 mutable-construction rule by @yuneng-berri in #43971
feat(vertex-ai): add vertex_ai/xai/grok-4.7 pricing by @berriai-litellm-provider-info-sync[bot] in #44059
feat(ui): drop the Beta badge from the Cost Optimization nav item by @devin-ai-integration[bot] in #43967
test(e2e): bill Sail windows that synchronous calls can still use by @yuneng-berri in #44058
ci(circleci): test Redis behavior against local Redis and print short tracebacks by @yuneng-berri in #44062
fix(router): keep silent_model out of embedding provider requests by @devin-ai-integration[bot] in #44064
fix(cost-map): add perplexity, openrouter, voyage and nebius models and fix registry metadata by @devin-ai-integration[bot] in #43907
feat(tool-policies): show the user who owns the key that discovered a tool by @devin-ai-integration[bot] in #43892
feat(proxy): gzip buffered responses for clients that accept it by @tin-berri in #44052
fix(auto-router): show actual and baseline spend for historical savings by @tin-berri in #44057
refactor(proxy): inject tracing receiver and access context by @yujonglee-berri in #44035
feat: improve trace ingestion and trace details by @yujonglee-berri in #43975
fix(proxy): enforce key/team vector_stores allowlist on /v1/rag/query by @devin-ai-integration[bot] in #43953
feat(lens): move traces and setup into Lens by @moe-berri in #44068
test(proxy-extras): run the db push timeout hint test without a database URL by @yuneng-berri in #44073
fix(proxy-extras): build the SpendLogs indexes in the migration job instead of in migrations by @devin-ai-integration[bot] in #43948
fix(mcp): resolve team-granted toolsets for non-admin keys and dashboard sessions by @devin-ai-integration[bot] in #43908
fix(bedrock): accept Converse messages with no content key by @devin-ai-integration[bot] in #43936
feat: add litellm.agent() to run claude code, codex, opencode and deep agents through the ai gateway by @ishaan-berri in #43885
feat(ui): add test trace, tracing key and otel endpoints to tracing setup by @ishaan-berri in #44090
feat(proxy): add LITELLM_DISABLE_LAZY_ROUTES to register optional routers at startup by @devin-ai-integration[bot] in #43911
test(e2e): keep 1ms-timeout deployments off the provider cache by @yuneng-berri in #44082
refactor(repositories): daily activity repository with centralized bounded usage queries by @devin-ai-integration[bot] in #43398
fix(bedrock): add beta for mid-conversation tool changes by @shrey-berri in #43833
feat(proxy): bounded daily activity routes (aggregated, search, model_top_keys, export, cache_leakage_keys) for all usage entities by @devin-ai-integration[bot] in #43408
feat(ui): usage pages consume bounded daily activity routes instead of storing all keys client-side by @devin-ai-integration[bot] in #43409
build(docker): drop the no-op PROXY_EXTRAS_SOURCE switch from the non-root image by @yuneng-berri in #44097
fix(proxy): persist SSO display name as user_alias on login by @devin-ai-integration[bot] in #44065
fix(cost-map): restore later azure Models API retirement dates and date gpt-6.1-sol by @berriai-litellm-provider-info-sync[bot] in #44072
feat(lens): simplify setup and investigation workflow by @moe-berri in #44089
chore(cost-map): sync openrouter prices from the models API by @berriai-litellm-provider-info-sync[bot] in #44105
fix(providers): keep thinking display updates beta by @shrey-berri in #43969
feat(rust): embed migration folders with a shared migrate! macro by @devin-ai-integration[bot] in #44104
refactor(tracing): normalize agent spans in Rust by @yujonglee-berri in #44071
perf(traces): recalculate ClickHouse TTL info only on retention changes by @devin-ai-integration[bot] in #44117
fix(proxy-extras): bound the lock waits of the partitioned SpendLogs index build by @devin-ai-integration[bot] in #44109
build(deps): bump oauthlib to 4.0.0 to clear osv-scan by @devin-ai-integration[bot] in #43899
fix(ui): label lens trace services as agents by @ishaan-berri in #44116
fix(ui): split the KeyActivityPanel condition chains to bring the lint budget back under its ceiling by @devin-ai-integration[bot] in #44114
chore: bump litellm-enterprise 0.1.72 -> 0.1.73, litellm-proxy-extras 0.4.103 -> 0.4.104 by @yuneng-berri in #44126

New Contributors

@galovics made their first contribution in #42949

Full Changelog: v1.105.0-dev.1...v1.105.0-dev.2

v1.103.2

2026-10-01T06:37:31Z

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:v1.103.2

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/v1.103.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.103.2

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

chore(release): sync stable/1.103.x to v1.103.1 by @yuneng-berri in #43824
fix(proxy): backport #40541, #43642, and #43656 to stable/1.103.x for v1.103.2 by @devin-ai-integration[bot] in #43897
fix(anthropic): backport #42152 and #42288 to stable/1.103.x by @devin-ai-integration[bot] in #43662
fix(proxy): backport #43962 to stable/1.103.x by @yuneng-berri in #43984

Full Changelog: v1.103.1...v1.103.2

v1.101.4

2026-10-01T05:10:19Z

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:v1.101.4

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/v1.101.4/cosign.pub \
  ghcr.io/berriai/litellm:v1.101.4

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

chore(release): sync stable/1.101.x to v1.101.3 by @yuneng-berri in #43822
chore(release): backport #41941, #41880, #43956 to stable/1.101.x and cut 1.101.4 by @yuneng-berri in #43943

Full Changelog: v1.101.3...v1.101.4

v1.105.0-dev.1

2026-09-30T08:44:29Z

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:v1.105.0-dev.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/v1.105.0-dev.1/cosign.pub \
  ghcr.io/berriai/litellm:v1.105.0-dev.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

test(e2e): cover Azure code_interpreter container files by native id with a service-account key by @devin-ai-integration[bot] in #43122
fix(router): await budget redis pipeline before sync reads (internal copy of #32618) by @devin-ai-integration[bot] in #43125
chore(cost-map): add openai cached image input prices from the pricing page by @berriai-litellm-provider-info-sync[bot] in #43143
feat(langfuse)!: migrate the sdk callback to langfuse v4 by @yucheng-berri in #36741
fix(key_management): count team unified access group MCP servers when validating key MCP grants by @devin-ai-integration[bot] in #41231
fix(ui): group cost optimization cache leakage by model group by @devin-ai-integration[bot] in #43008
fix(presidio): mask streamed /v1/messages output when the first upstream read is a keepalive, a data-less ping, or a split utf8 character by @devin-ai-integration[bot] in #43023
feat(mcp): share compatibility-aware result conversion across tool surfaces by @devin-ai-integration[bot] in #43089
fix(caching): propagate auth cache invalidation over Redis Cluster via a node-level pub/sub client by @devin-ai-integration[bot] in #43110
feat(model-catalog): add Rust registry validation by @devin-ai-integration[bot] in #43136
test(rust): reorganize core crate tests and split cache and OCR suites by @devin-ai-integration[bot] in #43177
fix(proxy): give SpendLogToolIndex its own share of the cleanup budget and log a per-run summary by @devin-ai-integration[bot] in #41768
feat(rust): hand upstream response headers to the native Messages stream by @devin-ai-integration[bot] in #43178
feat(testkit): agent clients for Claude Code, Codex and opencode by @devin-ai-integration[bot] in #43181
ci: run migrated unit selections on every event in legacy GHA shards by @yuneng-berri in #43182
test: move tests/test_litellm root and small trees into tests/unit by @yuneng-berri in #43186
fix(sso): gate /sso/debug routes behind ENABLE_SSO_DEBUG, off by default by @ojensen-berri in #43150
refactor(framer): replace Framer trait with tokio-util codecs by @devin-ai-integration[bot] in #43193
fix(anthropic): surface Responses bridge stream failures as Anthropic error events by @devin-ai-integration[bot] in #43126
test: move tests/test_litellm/llms into tests/unit/llms by @yuneng-berri in #43191
test: move tests/test_litellm integrations and secret_managers into tests/unit by @yuneng-berri in #43194
feat(mcp): configure protocol versions and capability discovery by @joshua-berri in #43169
fix(proxy): record streamed /v1/responses container ownership before the response.completed frame by @devin-ai-integration[bot] in #43140
fix(tests): drop the repeated UNIT_FLAG key in test_unit_shard_missing_paths by @devin-ai-integration[bot] in #43212
fix(router): count provider budget spend on every API surface by @daqiangganjun in #38172
fix(responses): fall back on pre-output stream drops, fail truncated streams, honor request_timeout by @devin-ai-integration[bot] in #43133
fix(sentry): scrub PII and secrets inside object reprs and nested locals, add SENTRY_SEND_DEFAULT_PII opt-in by @devin-ai-integration[bot] in #43123
fix(bedrock): surface a converse-stream 200 that decodes to no events as a 502 instead of an empty turn by @devin-ai-integration[bot] in #43213
fix(jwt): let x-litellm-team-id select DB membership teams when the token also carries a team claim by @devin-ai-integration[bot] in #43206
feat(integrations): add Databricks Zerobus trace logging callback by @devin-ai-integration[bot] in #42013
fix(vertex_ai): keep legacy bucket_name in credential resolution and add GCS_BATCH_BUCKET_NAME env var by @mubashir1osmani in #42803
feat(xai): add native xAI batches and files support by @devin-ai-integration[bot] in #42812
fix(proxy): send a real error event when a /v1/messages stream fails by @4refael in #41826
fix(logging): redact raw_request when turn_off_message_logging is set in the proxy config by @devin-ai-integration[bot] in #43219
fix(router): parse the classifier verdict out of surrounding prose instead of falling to the default tier by @devin-ai-integration[bot] in #43215
test(zerobus): move tests into active CI selection by @tin-berri in #43235
fix(proxy): keep the submitted body out of 422 validation errors by @devin-ai-integration[bot] in #43231
test: move tests/test_litellm core utils, routing, responses, caching and rust_bridge into tests/unit by @yuneng-berri in #43199
chore(cost-map): add fireworks us-only deepseek v4.1 flash priority prices by @berriai-litellm-provider-info-sync[bot] in #43247
chore(cost-map): sync openrouter prices and add perceptron-mk1.5 by @berriai-litellm-provider-info-sync[bot] in #43246
fix(cost-map): correct fireworks_ai deepseek-v4p1-flash pricing by @devin-ai-integration[bot] in #43253
feat(openrouter): add typesafe/jev-router to the cost map by @devin-ai-integration[bot] in #43248
chore(cost-map): add fireworks priority prices for muse glimmer 30b and deepseek v4 flash vision exp by @berriai-litellm-provider-info-sync[bot] in #43252
refactor(rust): move credential inheritance and SDK limits into a driver preflight by @devin-ai-integration[bot] in #43259
fix(router): hold Responses lifecycle events until output so a pre-output fallback announces one response by @devin-ai-integration[bot] in #43238
refactor(http): hand out an owned Client and route all providers through the pool by @devin-ai-integration[bot] in #43245
test: fix stale and state-leaking tests red on scheduled CircleCI by @yuneng-berri in #43266
fix(cost-map): retirement dates, chatgpt reasoning flags, bing pricing, bedrock mantle and mythos, azure gpt-5.6 alias, anthropic batch rates, new nebius, openrouter and xai rows by @devin-ai-integration[bot] in #42951
refactor(rust): promote anthropic messages out of experimental_pass_through by @devin-ai-integration[bot] in #43269
test: stop CI tests from downloading tokenizer files and images by @yuneng-berri in #43257
fix(rust): preserve nested optional import failures by @devin-ai-integration[bot] in #43265
ci: drop main and litellm_* branch filters from the CircleCI litellm-main workflows by @devin-ai-integration[bot] in #43272
feat(proxy): email alerts at configured percentages of a team member budget by @devin-ai-integration[bot] in #42665
fix(callbacks-legacy-python): traverse and release the retained headers dict by @devin-ai-integration[bot] in #43274
test(integration): port langfuse callbacks-in-db coverage to the local harness by @devin-ai-integration[bot] in #43282
test(e2e): accept the otel cost write as a linked root trace by @devin-ai-integration[bot] in #42931
test: finish the non-proxy half of tests/test_litellm by @devin-ai-integration[bot] in #43281
refactor(rust): prepare inference and auth foundations for the gateway by @devin-ai-integration[bot] in #43287
feat(rust): add config, router and gateway crates by @devin-ai-integration[bot] in #43289
fix(cost-map): remove duplicate openrouter/perceptron/perceptron-mk1.5 entry by @devin-ai-integration[bot] in #43273
fix(e2e): resolve the blank-S3 gateway repo root from the litellm package location by @devin-ai-integration[bot] in #42911
test(integration): run the Langfuse DB-callback test on its own scratch database by @yuneng-berri in #43288
refactor(rust): expand logging and test coverage across gateway and Anthropic messages by @devin-ai-integration[bot] in #43295
refactor: daily fresh tech debt cleanup, rolling PR (2026-09-25) by @devin-ai-integration[bot] in #43151
test(proxy_behavior): scope the management proxy fixture to its package so its spend monitor cannot race the spend tests by @devin-ai-integration[bot] in #43302
fix(cost-map): registry audit 2026-09-26, MAI-Image-2.5-Flash price, Databricks Claude Opus 5.5, Azure Foundry retirement dates by @devin-ai-integration[bot] in #43254
fix(cost-map): price fireworks deepseek v4.1 flash at the prices api value by @berriai-litellm-provider-info-sync[bot] in #43311
fix(ci): stop stale CI reds, keep unit tests off the host env, retry CyberArk policy conflicts by @yuneng-berri in #43294
fix(otel): detach post-response service spans by request phase, name redis spans by operation by @devin-ai-integration[bot] in #43237
test(e2e): assert only litellm-owned batch behavior and move the blank S3 env pin to an integration test by @yuneng-berri in #43321
test(integration): pin the team-admin status-code matrix across every management route by @ryan-crabbe-berri in #43249
feat(proxy): add fail_closed_rate_limit_enforcement to reject requests with 503 while Redis rate limit counters are unreachable by @devin-ai-integration[bot] in #43251
fix(responses): run stream failure and success hooks on the iterating loop instead of blocking it by @devin-ai-integration[bot] in #43270
fix(guardrails): block private destinations in custom code http_request and bound guardrail execution time by @devin-ai-integration[bot] in #43280
feat(sail): add Sail as a provider with service_tier mapped to its completion window by @devin-ai-integration[bot] in #42840
test(mcp): pin server resolution and authorization behavior by @joshua-berri in #43261
refactor(anthropic): rename experimental_pass_through to pass_through by @devin-ai-integration[bot] in #43329
refactor(mcp): add shared server resolver without changing callers by @joshua-berri in #43262
ci: fail on new unbounded SQL IN lists and add a Prisma chunking helper by @ryan-crabbe-berri in #42629
test(logging): drain the logging worker after each logging callback test so no later test inherits its events by @devin-ai-integration[bot] in #43344
feat(guardrails): scan retrieved vector store chunks with the request's pre-call guardrails by @devin-ai-integration[bot] in #43271
fix(mcp): adopt shared server resolution and caller authorization by @joshua-berri in #43263
fix(s3_v2): upload fresh events first, drop terminal failures and hour-old retries by default, opt-in adaptive concurrency by @devin-ai-integration[bot] in #43022
feat(proxy): add maximum_daily_tag_spend_retention_period cleanup setting by @devin-ai-integration[bot] in #39221
fix(cost-map): sync OpenRouter, Together, Cohere and Azure AI registry values with official sources by @devin-ai-integration[bot] in #43337
docs(pr-template): add the backport-stable label only for a P0 regression by @devin-ai-integration[bot] in #43351
fix(params): keep litellm* kwargs out of provider request bodies by construction by @shrey-berri in #43221
test: remove substring guard test_default_api_base by @devin-ai-integration[bot] in #43355
ci: cut CircleCI wall time without loosening test isolation by @yuneng-berri in #43347
fix(cost-map): correct azure gpt-4o-mini tts, transcribe, alias and MAI-Image-2.5 prices by @devin-ai-integration[bot] in #43357
fix(tests): stop VCR recording and replaying a test's own localhost upstream by @devin-ai-integration[bot] in #43346
test(integration): group /v1/messages contracts under tests/integration/messages_endpoint by @devin-ai-integration[bot] in #43352
fix(params): validate stream_chunk_size once, before any provider call by @shrey-berri in #43222
fix(openai): exclude fine-tuned and custom gpt-5-chat aliases from gpt-5 reasoning path by @devin-ai-integration[bot] in #43185
fix(langtrace): deliver spans to app.langtrace.ai/api/trace with x-api-key by @devin-ai-integration[bot] in #43322
fix(caching): stand default cache points down when extra_body hides a direct client mark by @yuneng-berri in #43341
fix(mcp): report reachability without stored credentials by @tin-berri in #43240
fix(mcp): align hub publication status and controls by @tin-berri in #43241
chore(cost-map): sync openrouter prices for deepseek, minimax, qwen and glm rows by @berriai-litellm-provider-info-sync[bot] in #43384
chore(cost-map): drop stale cache hit field from openrouter deepseek-v4-pro-0813 by @berriai-litellm-provider-info-sync[bot] in #43389
fix(responses): stream guardrail pre-call block as SSE with a typed output item by @devin-ai-integration[bot] in #42507
feat(e2e): read management routes back from the control plane replicas by @devin-ai-integration[bot] in #43373
feat(rust): add the openai_like chat config foundation by @devin-ai-integration[bot] in #43379
feat(router): opt in to prompt-cache cost routing by @tin-berri in #43232
feat(otel): add SigNoz preset for OpenTelemetry v2 by @devin-ai-integration[bot] in #43296
feat(cli): reuse saved agent setup and add reconfigure by @tin-berri in #43392
fix(streaming): keep litellm Usage on text-completion usage chunks by @yuneng-berri in #43047
fix(vertex_ai): consider tools when validating context caching min tokens by @agustin18 in #43319
fix(anthropic): drop thinking blocks with empty thinking text, not just missing signature by @shoemoney in #38049
test(e2e): report batch cleanup leftovers as a plain UserWarning by @yuneng-berri in #43405
fix(tools): salvage concatenated JSON tool call arguments by @kumarpriyanshu09 in #43260
fix(gemini): forward seed to the Gemini API instead of rejecting it by @YaseenBashaT in #43197
fix(vertex_ai): make Gemma fake streams work with traced Responses by @stewartpark in #43147
fix(responses): emit the reasoning item on streaming /v1/responses for signature-only thinking by @devin-ai-integration[bot] in #43414
fix(anthropic): forward the per-turn-control beta to Azure AI Foundry by @devin-ai-integration[bot] in #43415
test(vertex_ai): assert the usage the Gemma responses stream actually reports by @devin-ai-integration[bot] in #43422
fix(bedrock): keep the provider status code on unprocessable image errors by @devin-ai-integration[bot] in #43416
fix(vertex_ai): stop importing the vertexai SDK in partner-model completion by @anmolg1997 in #42274
fix(token_counter): count Gemini function_declarations tools by @devin-ai-integration[bot] in #43417
refactor(types): replace Any with proven types in 5 files by @devin-ai-integration[bot] in #43304
chore(cost-map): update azure_ai/grok-4.6 input price from Azure pricing page by @berriai-litellm-provider-info-sync[bot] in #43440
chore(cost-map): add azure_ai/MAI-Cyber-1-Flash by @berriai-litellm-provider-info-sync[bot] in #43446
refactor(rust): share anthropic types, request helpers, and streaming contracts across crates by @devin-ai-integration[bot] in #43426
feat(rust): expand gateway configuration parsing by @devin-ai-integration[bot] in #43460
refactor(rust): share call lifecycle across route-owned inference by @devin-ai-integration[bot] in #43461
feat(rust): add the HTTP host driver by @devin-ai-integration[bot] in #43462
build(rust): package the gateway container by @devin-ai-integration[bot] in #43471
refactor(rust): use shared execution in gateway inference by @devin-ai-integration[bot] in #43463
fix(cost-map): add deprecation_date to together_ai Salesforce/Llama-Rank-V1 by @berriai-litellm-provider-info-sync[bot] in #43507
feat(rust): support the HTTP Responses API by @devin-ai-integration[bot] in #43464
fix(cost-map): sync openrouter prices from the models API by @berriai-litellm-provider-info-sync[bot] in #43506
fix(cost-map): set together_ai gpt-oss-20b and gemma-4-31B-it deprecation_date to 2026-09-15 by @berriai-litellm-provider-info-sync[bot] in #43509
feat(rust): add litellm-db and litellm-db-testing workspace scaffolding by @devin-ai-integration[bot] in #43504
feat(rust): connect Python inference bindings to shared routes by @devin-ai-integration[bot] in #43465
feat(rust): add structured route lifecycle tracing by @devin-ai-integration[bot] in #43466
feat(rust): separate gateway authentication and authorization by @devin-ai-integration[bot] in #43467
feat(rust): add virtual key storage contracts by @devin-ai-integration[bot] in #43468
feat(rust): add gateway UI login and sessions by @devin-ai-integration[bot] in #43469
feat(rust): add the MCP gateway by @devin-ai-integration[bot] in #43470
refactor(rust): remove delivery routing abstraction by @devin-ai-integration[bot] in #43514
feat(azure): add mistral ocr pricing and azure max output limits by @berriai-litellm-provider-info-sync[bot] in #43530
refactor: clean up fresh tech debt from 2026-09-27 by @devin-ai-integration[bot] in #43538
refactor(types): replace Any with proven types in 8 files by @devin-ai-integration[bot] in #43551
test(rust): enforce shared upstream error contract in wheel checks by @devin-ai-integration[bot] in #43520
fix(proxy): log key owner identity on expired key auth failures by @devin-ai-integration[bot] in #43105
fix(cost-map): registry audit 2026-09-28, openai deep-research shutdown dates, azure deepseek v4.1 flash direct price, vertex gemini 3.8 live avatar price, drop azure_ai/muse-spark-1.3 by @devin-ai-integration[bot] in #43566
security(proxy): keep team callback credentials out of the stored request body by @devin-ai-integration[bot] in #43217
feat(model_prices): add claude-sonnet-5-5 model pricing by @krrish-berri-2 in #43586
test(integration): pin org-admin status codes in the team-admin matrix by @ryan-crabbe-berri in #43592
refactor(guardrails): fix agent 365 to the production endpoint and log the opt-in fail_open at error level by @devin-ai-integration[bot] in #43189
fix(model_prices): correct Claude Sonnet 5.5 capabilities and provider keys by @devin-ai-integration[bot] in #43587
test(unit): stop test modules from putting their own directory on sys.path by @yuneng-berri in #43421
refactor(rust): centralize host execution and compose callbacks by @devin-ai-integration[bot] in #43515
fix(proxy): unregister logging callbacks removed from the stored config by @yuneng-berri in #43428
chore(cost-map): take azure limits for deepseek-v4-flash-0731 and v3.2-speciale by @berriai-litellm-provider-info-sync[bot] in #43597
revert: "feat(proxy): server-side Team Usage export beyond the top-N key cap (#42996)" by @yassin-berriai in #43376
fix(cost-map): add web search flag and model page source to anthropic claude-sonnet-5-5 by @berriai-litellm-provider-info-sync[bot] in #43584
refactor(mcp): consolidate hub publication predicate by @tin-berri in #43394
fix(provider): accept 2xx status codes in unified_access_group create by @Louis-Vauterin in #42461
fix(panw_prisma_airs): honor experimental_use_latest_role_message_only on every request shape by @devin-ai-integration[bot] in #42447
fix(cost-map): add Vertex batch cache prices to vertex_ai/claude-sonnet-5-5 by @berriai-litellm-provider-info-sync[bot] in #43602
revert: "feat(usage): search team keys beyond the top-N in the Team usage view (#42857)" by @yassin-berriai in #43377
feat(providers): add Prism provider (internal copy of #40914) by @devin-ai-integration[bot] in #41961
test(integration): credential canary suite harness by @yucheng-berri in #43300
feat(bedrock): add xai grok-4.7 pricing and sync llama, mistral large 2407 and minimax m2.5 prices by @berriai-litellm-provider-info-sync[bot] in #43623
ci: sync the weekly release cycle with Linear releases by @yuneng-berri in #43636
feat(cache): select Rust caching through explicit cache objects by @devin-ai-integration[bot] in #43601
test(integration): credential canary slots for MCP and pass-through credentials by @yucheng-berri in #43308
test(integration): stored-config credential canary slots by @yucheng-berri in #43309
fix(guardrails): preserve Presidio output selection and restoration by @joshua-berri in #43401
fix(ui): rename All Models tab to Deployed Models and model filters to All Proxy Models by @devin-ai-integration[bot] in #43638
feat(mcp): scan and pin upstream tool descriptions by @devin-ai-integration[bot] in #43283
feat(cost-map): add bedrock_mantle rows for claude opus 5.5 and sonnet 5.5 by @devin-ai-integration[bot] in #43647
chore(codeowners): drop UI, migration, and CODEOWNERS self owners by @devin-ai-integration[bot] in #43653
feat(fireworks_ai): route and list the auto, auto-instant and firerouter routers by @devin-ai-integration[bot] in #43641
feat: add model leaderboard page by @ishaan-berri in #43649
fix(google_genai): preserve proxy_server_request in completion adapter by @hsm207 in #43536
fix(otel): send cache and reasoning tokens in langfuse usage_details by @ankit373 in #43553
fix(vertex_ai): forward the per-turn-control beta for per-message output_config by @DeviaVir in #43558
fix(cost-map): add tool calling and reasoning flags, correct max output for nebius DeepSeek-V4.1-Flash by @Flexomatic81 in #43588
fix(proxy): resolve model_group_alias in the zero-cost budget predicate by @fedaeho in #43512
refactor: clean up fresh tech debt from 2026-09-28 by @devin-ai-integration[bot] in #43674
refactor(types): replace Any with proven types in 7 files by @devin-ai-integration[bot] in #43704
docs(security): point readers to the security announcements mailing list signup by @devin-ai-integration[bot] in #43713
refactor(rust): add shared llms wire type derives by @devin-ai-integration[bot] in #43730
fix(router): stream /v1/messages lifecycle frames live when no fallback can take over by @devin-ai-integration[bot] in #43600
feat(cost-map): add baseten DeepSeek-V4.1-Flash-Fast by @berriai-litellm-provider-info-sync[bot] in #43735
feat(guardrails): send a configured gateway_name from noma_v2 to Noma by @itaimodi in #43678
chore(cost-map): add openai gpt-6.1-sol from the pricing page by @berriai-litellm-provider-info-sync[bot] in #43738
test(integration): callback credential canary slots C1-C3 and D5 by @yucheng-berri in #43630
test(integration): request-path credential canary slots D1-D4 by @yucheng-berri in #43307
test(integration): sweep proxy logs, metrics, a Datadog intake and the Logs drawer for credential canaries by @yucheng-berri in #43306
chore(cost-map): add azure and openrouter gpt-6.1-sol rows by @devin-ai-integration[bot] in #43744
test(proxy): classify every credential-bearing param for the canary suite by @yucheng-berri in #43298
fix(cost-map): lower fireworks up-to-4b size tier to the pricing page price by @berriai-litellm-provider-info-sync[bot] in #43740
fix(cost_calculator): bill chat per-second pricing once with a new cost_per_second field by @devin-ai-integration[bot] in #43614
chore(cost-map): add openai gpt-6-astra ultrafast tier prices from the pricing page by @berriai-litellm-provider-info-sync[bot] in #43745
fix(autorouter): compare historical and new savings consistently by @tin-berri in #43348
fix(model-prices): align Azure, Bedrock, Copilot, Gemini, Groq, OpenAI and OpenRouter entries with official docs by @devin-ai-integration[bot] in #43598
fix(guardrails): enable explicit PANW MCP output scanning by @devin-ai-integration[bot] in #43109
refactor(rust): orchestrate Messages route execution by @devin-ai-integration[bot] in #43719
fix(streaming): keep the served service_tier on streamed chunks and spend rows by @devin-ai-integration[bot] in #42870
feat(bedrock): add openai gpt-6.1-sol global and base rows by @berriai-litellm-provider-info-sync[bot] in #43758
perf(router): fetch cooldown state and usage counters in one Redis round trip by @devin-ai-integration[bot] in #43320
perf(proxy): hold one spend counter batch across admission and across post-call accounting by @devin-ai-integration[bot] in #43369
perf(responses): run aresponses through the async wrapper so the cache is read once by @devin-ai-integration[bot] in #43769
fix(router): bind Claude Code background sessions to their auto-router by @tin-berri in #43767
feat(bedrock): add openai.gpt-6.1-sol us geo cris and Mantle rows by @berriai-litellm-provider-info-sync[bot] in #43763
fix(proxy): recover session key owners from daily spend for usage attribution by @devin-ai-integration[bot] in #43642
chore(cost-map): take azure context limits from models-sold-directly by @berriai-litellm-provider-info-sync[bot] in #43759
perf(proxy): one request-scoped Redis pipeline for auth, spend, rate-limit and routing reads by @devin-ai-integration[bot] in #43407
fix(mcp): scope OpenAPI listings to the exact server prefix and drop upstream OAuth metadata when a server is saved by @devin-ai-integration[bot] in #43608
perf(proxy): one post-call Redis pipeline per backend for spend, rate-limit, routing and response-cache writes by @devin-ai-integration[bot] in #43779
fix(bedrock): set gpt-6.1-sol max output tokens to 131072 by @berriai-litellm-provider-info-sync[bot] in #43782
perf(proxy): refresh auth management objects through the request Redis pipeline by @devin-ai-integration[bot] in #43776
feat(agents): add identity storage and validation contracts by @joshua-berri in #43720
chore: bump litellm-enterprise 0.1.71 -> 0.1.72, litellm-proxy-extras 0.4.102 -> 0.4.103, litellm 1.104.0 -> 1.105.0 by @yuneng-berri in #43789
fix(auth): give UI/CLI session tokens their own AES-GCM context and header-safe shape by @yuneng-berri in #43790
chore(deps): bump pyjwt, moment and brace-expansion to clear osv-scan by @yuneng-berri in #43792
fix(ui): surface x-litellm-call-id in Logs search, table and drawer by @devin-ai-integration[bot] in #42436
fix(proxy): look up hashed key names with two spend log rows per key by @devin-ai-integration[bot] in #43656
fix(cost-map): add deprecation_date to two together_ai nvidia rows by @berriai-litellm-provider-info-sync[bot] in #43809
test(ci): refresh retired OpenAI tool-call models by @yuneng-berri in #43676
fix(ui): keep MCP permissions visible after key, team and MCP server saves by @devin-ai-integration[bot] in #43810
test(ci): repair MCP Responses and budget fixtures by @yuneng-berri in #43788
test(bedrock): accept regional aliases that inherit Converse routing by @yuneng-berri in #43785
fix(params): categorize internal param appropriately to prevent leaking into request by @shrey-berri in #43783
perf(router): honour the cooldown read interval in the routing prefetch by @devin-ai-integration[bot] in #43815

New Contributors

@daqiangganjun made their first contribution in #38172
@4refael made their first contribution in #41826
@agustin18 made their first contribution in #43319
@shoemoney made their first contribution in #38049
@YaseenBashaT made their first contribution in #43197
@stewartpark made their first contribution in #43147
@hsm207 made their first contribution in #43536
@ankit373 made their first contribution in #43553
@DeviaVir made their first contribution in #43558
@Flexomatic81 made their first contribution in #43588
@fedaeho made their first contribution in #43512

Full Changelog: v1.104.0-dev.2...v1.105.0-dev.1

v1.104.0-rc.2

2026-09-30T00:59:57Z

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:v1.104.0-rc.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/v1.104.0-rc.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.104.0-rc.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

Full Changelog: v1.104.0-rc.1...v1.104.0-rc.2

v1.103.1

2026-09-30T00:59:43Z

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:v1.103.1

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/v1.103.1/cosign.pub \
  ghcr.io/berriai/litellm:v1.103.1

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

Full Changelog: v1.103.0...v1.103.1

v1.102.2

2026-09-30T00:58:27Z

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:v1.102.2

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/v1.102.2/cosign.pub \
  ghcr.io/berriai/litellm:v1.102.2

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

Full Changelog: v1.102.1...v1.102.2

v1.101.3

2026-09-30T00:58:12Z

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:v1.101.3

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/v1.101.3/cosign.pub \
  ghcr.io/berriai/litellm:v1.101.3

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

Full Changelog: v1.101.2...v1.101.3

v1.100.4

2026-09-30T00:56:11Z

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:v1.100.4

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/v1.100.4/cosign.pub \
  ghcr.io/berriai/litellm:v1.100.4

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

Full Changelog: v1.100.3...v1.100.4

v1.104.0-rc.1

2026-09-28T08:01:07Z

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:v1.104.0-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/v1.104.0-rc.1/cosign.pub \
  ghcr.io/berriai/litellm:v1.104.0-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

test(e2e): cover Azure code_interpreter container files by native id with a service-account key by @devin-ai-integration[bot] in #43122
fix(router): await budget redis pipeline before sync reads (internal copy of #32618) by @devin-ai-integration[bot] in #43125
chore(cost-map): add openai cached image input prices from the pricing page by @berriai-litellm-provider-info-sync[bot] in #43143
feat(langfuse)!: migrate the sdk callback to langfuse v4 by @yucheng-berri in #36741
fix(key_management): count team unified access group MCP servers when validating key MCP grants by @devin-ai-integration[bot] in #41231
fix(ui): group cost optimization cache leakage by model group by @devin-ai-integration[bot] in #43008
fix(presidio): mask streamed /v1/messages output when the first upstream read is a keepalive, a data-less ping, or a split utf8 character by @devin-ai-integration[bot] in #43023
test: backport stale and state-leaking test fixes to rc/1.104.0 (#43266) by @yuneng-berri in #43267
chore: backport CI fixes, #43206 and Rust changes to rc/1.104.0 by @yuneng-berri in #43338
fix(s3_v2): backport #43022 to rc/1.104.0 by @yuneng-berri in #43354
chore: rebuild Admin UI bundle for rc/1.104.0 by @yuneng-berri in #43372
chore: backport CircleCI speedups (#43347) and cost-map fixes (#42951) to rc/1.104.0 by @yuneng-berri in #43366
revert(usage): remove the top-N key cap, its follow-ups and the daily global spend rollup code from rc/1.104.0 by @yuneng-berri in #43385
fix(streaming): backport text-completion usage chunk fix (#43047) to rc/1.104.0 by @yuneng-berri in #43402
test(e2e): report batch cleanup leftovers as a plain UserWarning on rc/1.104.0 by @yuneng-berri in #43406
test(e2e): skip the LangSmith batch serialization e2e on rc/1.104.0 until CI has a LangSmith key by @yuneng-berri in #43418
test(unit): stop test modules from putting their own directory on sys.path on rc/1.104.0 by @yuneng-berri in #43420
fix(proxy): unregister logging callbacks removed from the stored config on rc/1.104.0 by @yuneng-berri in #43432

Full Changelog: v1.104.0-dev.2...v1.104.0-rc.1