BerriAI/litellm v1.106.0-dev.3 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:v1.106.0-dev.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/v1.106.0-dev.3/cosign.pub \
  ghcr.io/berriai/litellm:v1.106.0-dev.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

What's Changed

test(integration): boot owned proxies with a readiness-sized worker healthcheck budget by @devin-ai-integration[bot] in #45292
fix(utils): make supports_audio_output read the supports_audio_output cost-map key by @devin-ai-integration[bot] in #45294
fix(lens): refresh runs until gateway costs are complete by @tin-berri in #45105
fix(lint-gates): fail on an empty base scan instead of blaming the change for every violation by @devin-ai-integration[bot] in #45314
refactor(proxy): type fresh Moyai connect helpers and drop MCP rpm getattr by @devin-ai-integration[bot] in #45315
revert(lint-gates): accept an empty base scan again, since zero violations is a legitimate count by @devin-ai-integration[bot] in #45319
fix(bedrock_mantle): send OpenAI explicit prompt cache breakpoints for GPT-5.6 and newer by @msdianprince-7 in #38729
test: make 77 legacy live tests offline in litellm_utils, router_unit and responses dirs by @devin-ai-integration[bot] in #45298
test: move 81 legacy live tests in pass-through, spend, batches, audio, search, guardrails, image and ocr dirs offline by @devin-ai-integration[bot] in #45288
feat(ui): add TypeSafe and Strands Decider to the Add Model provider list by @devin-ai-integration[bot] in #45335
feat(mcp): support upstream OAuth client metadata identities by @joshua-berri in #45231
fix(proxy-extras): name the database error when the Lens rename check cannot run by @devin-ai-integration[bot] in #45341
test(integration): run the response cache tool-call tests on a proxy without the message cap by @devin-ai-integration[bot] in #45350
fix(logging): skip sync success callbacks for internal sub-calls, deflake RAG and Langfuse tests by @devin-ai-integration[bot] in #45345
fix(azure): move sora-2 retirement date to the later Models API date by @berriai-litellm-provider-info-sync[bot] in #45355
fix(lens): block teamless feedback writes and fix retention test by @ishaan-berri in #45266
fix(ci): trim rust-test debuginfo so the job fits on the runner disk by @devin-ai-integration[bot] in #45318
fix(ui): list only catalog models in the Add Model picker by @devin-ai-integration[bot] in #44877
feat(rust): add standalone typed LLM wire contracts by @yujonglee-berri in #45188
test: delete 34 legacy tests already covered by e2e by @devin-ai-integration[bot] in #45340
fix(responses): keep tool_result next to tool_use on Anthropic previous_response_id continuations by @mateo-berri in #45322
test: realign stale decisions and lens tests with merged behaviour by @devin-ai-integration[bot] in #45346
feat(prometheus): expose per-project per-model rate limit allowed and used gauges by @devin-ai-integration[bot] in #43561
fix(mcp): preserve elicitation context and report relay failures by @joshua-berri in #45255
docs(rust): add ADRs for the Rust core by @nate-berri in #45205
test: move 76 live top-level tests to offline unit and integration coverage by @devin-ai-integration[bot] in #45352
fix(proxy): mark background responses stale_expired when provider returns 404 by @devin-ai-integration[bot] in #41724
fix(bedrock): take context, output limits and EOL dates from the Bedrock model cards by @berriai-litellm-provider-info-sync[bot] in #45409
chore(cost-map): add openai gpt-6.1-sol ultrafast tier prices from the pricing page by @berriai-litellm-provider-info-sync[bot] in #45408
fix(bedrock_mantle): take gpt-oss output limits and EOL dates from the Bedrock model cards by @berriai-litellm-provider-info-sync[bot] in #45417
fix(rust): path-qualify attribute aliases so rust-analyzer resolves them by @nate-berri in #45416
fix(proxy): use budget reset window for projected spend alerts by @Solomon-mithra in #31942
test(proxy): audit the websocket rejection log on every live route by @mateo-berri in #45293
fix(proxy): retry lock-timed-out daily spend batches in place so the shutdown flush keeps them by @mateo-berri in #45273
fix(vertex_ai): forward the inline-tools-2026-09-15 beta to Vertex and Anthropic by @devin-ai-integration[bot] in #45297
feat(fireworks_ai): forward the LiteLLM user id as user behind fireworks_forward_user_id by @yucheng-berri in #45265
fix(model_prices): add Bedrock flex prices, Anthropic web search flags, Gemini shutdown dates, OpenRouter alias drift, Mistral Large 4 context by @devin-ai-integration[bot] in #44910
feat(voyage): rebrand to VoyageAI by MongoDB and route MongoDB keys to ai.mongodb.com by @fzowl in #41812
feat(gemini): accept file content blocks with video_metadata on multimodal embeddings by @devin-ai-integration[bot] in #45305
feat(rust): add the Anthropic beta header policy by @yujonglee-berri in #45419
test(decisions): post System One bodies to /v1/systemone in the translation bases by @yuneng-berri in #45327
fix(ci): drop the publicly known master key prefix from the voyage routing test by @devin-ai-integration[bot] in #45431
feat(guardrails): per-mode stream_scope with bedrock stream and pass-through fixes by @devin-ai-integration[bot] in #43801
refactor(python-bridge): take NativeCall directly and fold routes into per-route folders by @yujonglee-berri in #45413
fix(lens): prevent progress updates from starving analysis budget reservations by @moe-berri in #45432
refactor(rust): derive strum VariantArray and string conversions by @yujonglee-berri in #45434
chore(decisions): remove the System One converters and OpenAI spec types left dead by #45214 by @devin-ai-integration[bot] in #45442
feat(spend_logs): configure which metadata fields are stored in LiteLLM_SpendLogs by @devin-ai-integration[bot] in #44659
fix(ui): show the Add Model picker once the model catalog loads after a provider is picked by @yuneng-berri in #45426
fix(rust): add the inline-tools-2026-09-15 beta to AnthropicBeta by @devin-ai-integration[bot] in #45439
fix(bedrock): count tokens on bedrock-mantle when bedrock-runtime cannot count a Claude model by @devin-ai-integration[bot] in #45317
fix(ui): link model access group chips to the access group filter by @devin-ai-integration[bot] in #45402
fix(proxy): accept team-scoped models by their public name on POST /fallback by @devin-ai-integration[bot] in #45455
fix(otel): emit OpenInference tool calls and metadata on Arize OTel v2 spans by @devin-ai-integration[bot] in #43698
test(ui-e2e): update usage page selectors after the #45221 redesign by @devin-ai-integration[bot] in #45347
test: replace 61 live logging and otel tests with offline unit and integration coverage by @devin-ai-integration[bot] in #45362
fix(bedrock): remove the bare xai.grok-4.7 cost-map row that AWS cannot invoke by @devin-ai-integration[bot] in #45460
test: make 38 legacy live base-class translation tests offline by @devin-ai-integration[bot] in #45351
fix(bedrock): split the tag for gpt-oss only on native Chat Completions by @devin-ai-integration[bot] in #44432
feat(ui): add Cmd+K command palette with key search by @ishaan-berri in #45456
test(streaming): expect the unwrapped provider error in bridged /v1/messages error frames by @devin-ai-integration[bot] in #45465
refactor(python-bridge): derive NativeCall extraction by @yujonglee-berri in #45449
refactor(python-bridge): ship a signature base and read the resolved call by @yujonglee-berri in #45450
fix(cost-map): update baseten DeepSeek-V4.1-Flash-Fast cache read price and max output by @berriai-litellm-provider-info-sync[bot] in #45468
ci(unit): add unit passed collector job and fold proxy-db shards into test-unit.yml by @devin-ai-integration[bot] in #45466
feat(guardrails): support logging_only mode for the Akto guardrail by @yucheng-berri in #45461
fix(router): drop the encrypted reasoning a fallback hop's target cannot decrypt by @devin-ai-integration[bot] in #45393
chore(cost-map): sync openrouter prices from the models API by @berriai-litellm-provider-info-sync[bot] in #45469
fix(cli): pin pi compat flags so lite pi stops sending store to Anthropic models by @devin-ai-integration[bot] in #40739
feat(bedrock): add twelvelabs pegasus 1.5 inference profiles by @berriai-litellm-provider-info-sync[bot] in #45477
feat(sdk): build core from an independent packaging manifest by @joshua-berri in #44340
fix(proxy): save file details for every batch output file so they list and retrieve by @devin-ai-integration[bot] in #41761
fix(bedrock): remove the bare openai.gpt-6.1-sol cost-map row that AWS cannot invoke by @devin-ai-integration[bot] in #45479
refactor(sdk): separate core AWS and tokenizer dependencies by @joshua-berri in #44447
fix(router): match deployment pricing ids against the cost map only within the deployment's provider by @devin-ai-integration[bot] in #45472
fix(bedrock): add gpt-6.1-sol ultrafast tier prices from the Bedrock model card by @berriai-litellm-provider-info-sync[bot] in #45482
fix(ci): import seed_tracing_fixtures from the pytest scripts path in rust trace tests by @devin-ai-integration[bot] in #45485
fix(bedrock): take gpt-6.1-sol context window from the Bedrock model card by @berriai-litellm-provider-info-sync[bot] in #45488
feat(ui): add the evaluation mode to the Add Model form by @devin-ai-integration[bot] in #45481
feat(mcp): translate input requests and bind continuations by @joshua-berri in #45464
fix(bedrock): add the claude-sonnet-4-5 EOL date from the Bedrock model card by @berriai-litellm-provider-info-sync[bot] in #45496
fix(token_counter): price a base64 PDF document per page instead of as one image by @devin-ai-integration[bot] in #45301
ci: render lint, unit and smoke checks as / with one collector per tier by @devin-ai-integration[bot] in #45480
test(routing): wait on deployment registration instead of boot model-info traffic by @yuneng-berri in #45267
fix(anthropic): count a leading system run through count_tokens' system parameter by @devin-ai-integration[bot] in #45463
ci: run only integration jobs on PR CircleCI pipelines and move router and guardrails suites to GHA by @devin-ai-integration[bot] in #45509
test: update stale MCP call_tool and usage card assertions, add timeout headroom to request-log index boot test by @devin-ai-integration[bot] in #45523
feat(credentials): add display_name and make credential_name immutable on PATCH by @devin-ai-integration[bot] in #43148
fix(ui): let the model picker remove selections that are no longer available by @yuneng-berri in #43944
refactor(proxy): inject one UTC clock read per operation into gateway tracking, PTU rollup and Mavvrik export by @devin-ai-integration[bot] in #45520
refactor(proxy): inject the clock into the v1 parallel request limiter and pin it in its tests by @devin-ai-integration[bot] in #45522
chore: bump litellm-enterprise 0.1.75 -> 0.1.76, litellm-proxy-extras 0.4.107 -> 0.4.108 by @devin-ai-integration[bot] in #45534
feat(ui): configure OpenAI workload identity federation from the LLM Credentials and Add Model forms by @mateo-berri in #45528

New Contributors

@msdianprince-7 made their first contribution in #38729
@Solomon-mithra made their first contribution in #31942

Full Changelog: v1.106.0-dev.2...v1.106.0-dev.3