Skip to main content
LiteLLM is moving to Rust Read the latest updates.

1.105.0rc1 - Claude Sonnet 5.5, GPT-6.1 Sol, litellm.agent(), Lens & Agent Traces

Deploy this version​

docker run \
-e LITELLM_MASTER_KEY=sk-<paste-a-long-random-key> \
-e DATABASE_URL=postgresql://<user>:<password>@<host>:5432/<dbname> \
-e STORE_MODEL_IN_DB=True \
-p 4000:4000 \
docker.litellm.ai/berriai/litellm:1.105.0-rc.1

The published GitHub tag is v1.105.0-rc.1. These notes compare it with v1.104.0-rc.1, the previous release candidate cut from main. Changes backported onto rc/1.104.0 and already shipped in v1.104.0 are omitted

Customer-facing changes come first. Test, CI and internal changes are listed at the bottom

Breaking Changes

These callouts cover user-facing behavior that differs from v1.104.0, the latest stable release

ENFORCE_PRISMA_MIGRATION_CHECK=false is now ignored, so a proxy whose migrations fail will not start. This keeps a new version from serving traffic against an outdated database schema and failing requests. See the v1.104.0 breaking changes and PR #44141

Bedrock GPT-5.6, GPT-6 and GPT-6.1 move from Converse to native Chat Completions, so response ids, service_tier and reasoning fields change shape. Use bedrock/converse/<model> to stay on Converse. See PR #44307

/sso/debug/* returns 404 unless ENABLE_SSO_DEBUG=true. See PR #43150

Key Highlights​

  • New models on day one: Claude Sonnet 5.5 across Anthropic, Bedrock, Bedrock Mantle, Vertex AI, Azure AI, OpenRouter and Perplexity, GPT-6.1 Sol across OpenAI, Azure, Bedrock and OpenRouter, and Grok 4.7 on Bedrock and Vertex AI, among 78 new catalog entries
  • New providers and routes: Prism, Sail and Cortecs providers, native xAI batches and files, Fireworks router models, and Bedrock GPT-5.6+ served on native Chat Completions
  • litellm.agent(): run Claude Code, Codex, OpenCode and Deep Agents through the AI gateway from the SDK
  • Lens and Agent Traces: OTLP trace ingestion stored in ClickHouse with matched spend, scoped SQL over traces, a chat-style run view, and Lens (Beta) investigations run by a separate worker
  • Agent identities: register agents with Entra ID identities, authenticate delegated requests and enforce authoritative agent permissions
  • Faster proxy at scale: one request-scoped Redis pipeline for auth, spend, rate-limit and routing reads, gzip for buffered responses, and usage pages that page keys from the server instead of loading every key into the browser

New Providers and Endpoints​

New Providers (3 new providers)​

ProviderSupported LiteLLM EndpointsDescription
Prism/v1/chat/completions, /v1/responses, /v1/messagesPrism inference as an OpenAI-compatible provider, with DeepSeek V4 Flash and V4.1 Flash in the catalog
Sail/v1/chat/completions, /v1/responses, /v1/messages12 Sail models, with service_tier mapped to Sail's completion window and billed at that window's price, including a new balanced tier
Cortecs/v1/chat/completions, /v1/responses, /v1/messagesCortecs, an EU LLM router, as an OpenAI-compatible provider

Expanded provider endpoint support​

ProviderEndpointWhat you can do
xAI/v1/files, /v1/batchesRun native xAI batches and files
Amazon Bedrock/v1/chat/completionsServe GPT-5.6, GPT-6 and GPT-6.1 on bedrock-runtime's native Chat Completions, with a chat_completions/ opt-in for gpt-oss and Grok
Fireworks AI/v1/chat/completionsRoute to and list the auto, auto-instant and firerouter routers

New Models / Updated Models​

New Model Support (78 new models)​

Counts represent new catalog identifiers compared with v1.104.0, including aliases and regional variants. Prices below are the values bundled in this release, in USD; runtime pricing-map reloads can update them. Input and output columns show base token rates; long-context, cache, image-token, and other specialized rates depend on the model

ProviderModelContext WindowInput ($/1M tokens)Output ($/1M tokens)Features / special pricing
Amazon Bedrockanthropic.claude-sonnet-5-51,000,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockapac.anthropic.claude-sonnet-5-51,000,000$2.2$11Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockau.anthropic.claude-sonnet-5-51,000,000$2.2$11Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockbedrock/us-gov-east-1/anthropic.claude-sonnet-5-51,000,000$2.4$12Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockbedrock/us-gov-west-1/anthropic.claude-sonnet-5-51,000,000$2.4$12Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockeu.anthropic.claude-sonnet-5-51,000,000$2.2$11Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockglobal.anthropic.claude-sonnet-5-51,000,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockglobal.openai.gpt-6.1-sol1,050,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching
Amazon Bedrockglobal.xai.grok-4.7500,000$2$6Chat; Reasoning; Vision; Tool calling
Amazon Bedrockjp.anthropic.claude-sonnet-5-51,000,000$2.2$11Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockopenai.gpt-6.1-sol1,050,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching
Amazon Bedrockus-gov.anthropic.claude-sonnet-5-51,000,000$2.4$12Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockus.anthropic.claude-sonnet-5-51,000,000$2.2$11Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Amazon Bedrockus.openai.gpt-6.1-sol1,050,000$2.2$11Chat; Reasoning; Vision; Tool calling; Prompt caching
Amazon Bedrockus.xai.grok-4.7500,000$2.2$6.6Chat; Reasoning; Vision; Tool calling
Amazon Bedrockxai.grok-4.7500,000$2$6Chat; Reasoning; Vision; Tool calling
Anthropicclaude-sonnet-5-51,000,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search; Computer use
Azure AIazure_ai/MAI-Cyber-1-Flash256,000$0.6$3.5Chat; Reasoning; Tool calling; Prompt caching
Azure AIazure_ai/claude-sonnet-5-51,000,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Azure AIazure_ai/gpt-6.1-sol922,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search; Computer use
Azure AIazure_ai/mistral-ocr-2505---OCR; ocr_cost_per_page: $0.001
Azure AIazure_ai/mistral-ocr-2512---OCR; ocr_cost_per_page: $0.002
Azure OpenAIazure/gpt-6.1-sol922,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search; Computer use
Azure OpenAIazure/gpt-6.1-sol-2026-09-29922,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search; Computer use
Basetenbaseten/deepseek-ai/DeepSeek-V4.1-Flash-Fast1,048,576$0.6$2.4Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Bedrock Mantlebedrock_mantle/anthropic.claude-opus-5-51,000,000$4.4$22Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Bedrock Mantlebedrock_mantle/anthropic.claude-sonnet-5-51,000,000$2.2$11Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Bedrock Mantlebedrock_mantle/openai.gpt-6.1-sol1,050,000$2.2$11Responses; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Bedrock Mantlebedrock_mantle/us-gov-west-1/anthropic.claude-opus-5-51,000,000$4.8$24Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Bedrock Mantlebedrock_mantle/us-gov-west-1/anthropic.claude-sonnet-5-51,000,000$2.4$12Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Databricksdatabricks/databricks-claude-opus-5-51,000,000$4.00001$19.99998Chat; Reasoning; Vision; Tool calling; Prompt caching
Fireworks AIfireworks_ai/accounts/fireworks/routers/auto---Chat; Reasoning; Tool calling; Structured output
Fireworks AIfireworks_ai/accounts/fireworks/routers/auto-instant---Chat; Reasoning; Tool calling; Structured output
Fireworks AIfireworks_ai/accounts/fireworks/routers/firerouter---Chat; Reasoning; Tool calling; Structured output
Nebiusnebius/Qwen/Qwen3.8-27B262,144$0.45$3Chat; Reasoning; Tool calling
OpenAIgpt-4o-mini-tts-2025-03-20-$0.6$10Speech; output_cost_per_second: $0.00025
OpenAIgpt-6.1-sol922,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search; Computer use
OpenRouteropenrouter/anthropic/claude-sonnet-5.51,000,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search
OpenRouteropenrouter/anthropic/claude-sonnet-5.5:batch1,000,000$1$5Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search
OpenRouteropenrouter/apodex/apodex-1.1-mini:free262,144$0$0Chat; Reasoning; Tool calling; Structured output
OpenRouteropenrouter/openai/gpt-6.1-sol1,050,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search
OpenRouteropenrouter/openai/gpt-6.1-sol-pro1,050,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search
OpenRouteropenrouter/openai/gpt-6.1-sol-pro:batch1,050,000$1$5Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search
OpenRouteropenrouter/openai/gpt-6.1-sol:batch1,050,000$1$5Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Web search
OpenRouteropenrouter/typesafe/jev-router1,000,000$0$0Chat; Reasoning; Vision; Tool calling; Structured output; PDF input; Audio input
OpenRouteropenrouter/unbiased/pareto-26.10-preview1,048,576$0.8$3.2Chat; Vision; Tool calling; Prompt caching
Perplexityperplexity/anthropic/claude-fable-5-1-$10$50Responses
Perplexityperplexity/anthropic/claude-opus-5-5-$4$20Responses
Perplexityperplexity/anthropic/claude-sonnet-5-5-$2$10Responses; Tool calling; Web search
Perplexityperplexity/google/gemini-3.8-flash-$0.75$3.75Responses
Perplexityperplexity/openai/gpt-6-luna-$0.1$0.5Responses
Perplexityperplexity/openai/gpt-6-sol-$2$10Responses
Perplexityperplexity/openai/gpt-6.1-sol-$2$10Responses
Perplexityperplexity/xai/grok-4.7-$2$6Responses
Prismprism/deepseek-v4-flash1,000,000$0.17$0.21Chat; Reasoning; Tool calling; Prompt caching; Structured output
Prismprism/deepseek-v4.1-flash1,000,000$0.17$0.63Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Sailsail/Qwen/Qwen3.6-35B-A3B262,144$0.05$0.4Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Sailsail/deepseek-ai/DeepSeek-V4-Flash-07311,048,576$0.09$0.18Chat; Reasoning; Tool calling; Prompt caching; Structured output
Sailsail/deepseek-ai/DeepSeek-V4-Pro-08131,048,576$0.92$2.77Chat; Reasoning; Tool calling; Prompt caching; Structured output
Sailsail/deepseek-ai/DeepSeek-V4.1-Flash1,048,576$0.15$0.6Chat; Reasoning; Tool calling; Prompt caching; Structured output
Sailsail/google/gemma-4-12B-it16,384$0.3$2Chat; Reasoning; Tool calling; Prompt caching; Structured output
Sailsail/google/gemma-4-31B-it256,000$0.4$0.6Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Sailsail/moonshotai/Kimi-K2.6262,144$1$4Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Sailsail/moonshotai/Kimi-K31,048,576$2.5$12.5Chat; Reasoning; Tool calling; Prompt caching; Structured output
Sailsail/nvidia/Gemma-4-31B-IT-NVFP4262,144$0.14$0.4Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Sailsail/openai/gpt-oss-120b131,072$0.06$0.4Chat; Reasoning; Tool calling; Prompt caching; Structured output
Sailsail/zai-org/GLM-5.31,048,576$0.98$3.08Chat; Reasoning; Tool calling; Prompt caching; Structured output
Sailsail/zai-org/GLM-5.3-Flash1,048,576$0.11$0.35Chat; Reasoning; Tool calling; Prompt caching; Structured output
Together AItogether_ai/Salesforce/Llama-Rank-V18,192$0.1$0Rerank
Together AItogether_ai/meta-llama/Meta-Llama-3.1-8B16,384$0.2$0.2Completion
Vertex AIvertex_ai/claude-sonnet-5-51,000,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Vertex AIvertex_ai/claude-sonnet-5-5@default1,000,000$2$10Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output; PDF input; Computer use
Vertex AIvertex_ai/gemini-3.8-flash-lite-tts8,192$0.5$6Speech
Vertex AIvertex_ai/gemini-3.8-flash-tts8,192$0.5$9Speech
Vertex AIvertex_ai/xai/grok-4.7524,288$2$6Chat; Reasoning; Vision; Tool calling; Prompt caching; Structured output
Voyage AIvoyage/rerank-18,000$0.05$0Rerank
Voyage AIvoyage/rerank-lite-14,000$0.02$0Rerank
Voyage AIvoyage/voyage-large-2-instruct16,000$0.12$0Embedding

Updated pricing (59 models)​

Provider / modelChanged token prices (USD per 1M tokens)
azure/eu/gpt-6-astraInput: $11 to $12; Output: $55 to $60; Cache read: $1.1 to $1.2; Cache write: $13.75 to $15
azure/gpt-4o-miniInput: $0.165 to $0.15; Output: $0.66 to $0.6
azure/gpt-4o-mini-ttsInput: $2.5 to $0.6
azure_ai/deepseek-v4.1-flashInput: $0.375 to $0.3; Output: $1.5 to $1.2; Cache read: $0.008 to $0.006
azure_ai/grok-4.6Input: $2 to $1.25
bedrock/ap-southeast-2/minimax.minimax-m2.5Input: $0.309 to $0.31; Output: $1.236 to $1.24
fireworks-ai-up-to-4bInput: $0.2 to $0.1; Output: $0.2 to $0.1
fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flashInput: $0.22 to $0.3; Output: $0.66 to $1.2; Cache read: $0.007 to $0.006
fireworks_ai/deepseek-v4p1-flashInput: $0.22 to $0.3; Output: $0.66 to $1.2; Cache read: $0.007 to $0.006
github_copilot/claude-haiku-4.5Input: not set to $1; Output: not set to $5; Cache read: not set to $0.1; Cache write: not set to $1.25
github_copilot/claude-sonnet-4Input: not set to $3; Output: not set to $15; Cache read: not set to $0.3; Cache write: not set to $3.75
github_copilot/gpt-5-miniInput: not set to $0.25; Output: not set to $2; Cache read: not set to $0.025
github_copilot/gpt-5.3-codexInput: not set to $1.75; Output: not set to $14; Cache read: not set to $0.175
meta.llama3-1-405b-instruct-v1:0Input: $5.32 to $2.4; Output: $16 to $2.4
meta.llama3-1-70b-instruct-v1:0Input: $0.99 to $0.72; Output: $0.99 to $0.72
meta.llama3-2-11b-instruct-v1:0Input: $0.35 to $0.16; Output: $0.35 to $0.16
meta.llama3-2-90b-instruct-v1:0Input: $2 to $0.72; Output: $2 to $0.72
mistral.mistral-large-2407-v1:0Input: $3 to $2; Output: $9 to $6
moonshotai.kimi-k3Input: $3 to $3.3; Output: $15 to $16.5; Cache read: $0.3 to $0.33; Cache write: $3.75 to $4.125
openrouter/deepseek/deepseek-chatInput: $0.32 to $0.2574; Output: $0.89 to $1.0287
openrouter/deepseek/deepseek-chat-v3-0324Input: $0.25 to $0.29; Output: $1 to $1.14; Cache read: not set to $0.11
openrouter/deepseek/deepseek-v3.1-terminusInput: $0.27 to $0.3
openrouter/deepseek/deepseek-v3.2Input: $0.269 to $0.28; Output: $0.4 to $0.42; Cache read: $0.1345 to $0.028
openrouter/deepseek/deepseek-v4-flashInput: $0.049 to $0.04186; Output: $0.098 to $0.08372; Cache read: $0.0098 to $0.008372
openrouter/deepseek/deepseek-v4-flash-0731Input: $0.03 to $0.0108; Output: $0.32 to $1.28; Cache read: $0.016 to $0.0108
openrouter/deepseek/deepseek-v4-flash-vision-expInput: $0.22 to $0.2156; Output: $0.66 to $0.6468; Cache read: $0.007 to $0.00686
openrouter/deepseek/deepseek-v4-proInput: $0.844944 to $0.2088; Output: $1.689888 to $0.4176; Cache read: $0.070412 to $0.0174
openrouter/deepseek/deepseek-v4-pro-0813Input: $0.462 to $1.32; Output: $1.386 to $3.96; Cache read: $0.0154 to $0.044
openrouter/deepseek/deepseek-v4.1-flashInput: $0.3 to $0.03; Output: $1.2 to $0.5; Cache read: $0.006 to $0.01
openrouter/google/gemma-4-26b-a4b-itInput: $0.09 to $0.0765; Output: $0.3 to $0.255; Cache read: $0.05 to $0.0425
openrouter/inclusionai/ling-3.0-flash-vlInput: $0.06 to $0.021; Output: $0.18 to $0.0616; Cache read: $0.012 to $0.0042
openrouter/meta/muse-glimmer-30bInput: $0.3 to $0.35; Output: $1.2 to $1.5
openrouter/minimax/minimax-m1Input: $0.4 to $0.55
openrouter/minimax/minimax-m2.7Input: $0.3 to $0.21; Output: $1.2 to $0.84; Cache read: $0.06 to $0.042
openrouter/moonshotai/kimi-k2.6Input: $0.95 to $0.43415; Output: $4 to $1.828; Cache read: $0.16 to $0.07312
openrouter/moonshotai/kimi-k2.7-codeInput: $0.6562 to $0.6712; Output: $3.3 to $3.35
openrouter/moonshotai/kimi-k3Input: $3 to $0.4357; Output: $15 to $10; Cache read: $0.3 to $0.4357
openrouter/nvidia/nemotron-3.5-lightningInput: $0.08 to $0.0595; Output: $0.2 to $0.17; Cache read: $0.04 to $0.02975
openrouter/openai/gpt-5.6-sol-proInput: $2 to $4; Output: $10 to $20; Cache read: $0.2 to $0.4; Cache write: $2.5 to $5
openrouter/openai/gpt-oss-120bInput: $0.15 to $0.037; Output: $0.6 to $0.17; Cache read: $0.075 to not set
openrouter/openai/gpt-oss-20bCache read: $0.03 to $0.009
openrouter/prism-ml/ternary-bonsai-2-27bCache read: not set to $0.0375
openrouter/qwen/qwen3-30b-a3b-instruct-2507Input: $0.1 to $0.04815; Output: $0.3 to $0.19305
openrouter/qwen/qwen3-vl-30b-a3b-instructInput: $0.13 to $0.15; Output: $0.52 to $0.6
openrouter/qwen/qwen3.5-35b-a3bInput: $0.3125 to $0.1625; Output: $1.25 to $1.3
openrouter/qwen/qwen3.6-27bOutput: $2.7 to $3.2
openrouter/x-ai/grok-4.7Input: $1.6 to $2; Output: $4.8 to $6; Cache read: $0.4 to $0.5
openrouter/z-ai/glm-4.6vCache read: $0.055 to $0.05
openrouter/z-ai/glm-4.7Input: $0.4 to $0.6; Output: $1.75 to $2.2; Cache read: $0.08 to $0.11
openrouter/z-ai/glm-5.1Input: $0.966 to $0.9646; Output: $3.036 to $3.0316; Cache read: $0.1794 to $0.17914
openrouter/z-ai/glm-5.2Input: $0.6496 to $0.41; Output: $2.0416 to $3.99; Cache read: $0.12064 to $0.26
openrouter/z-ai/glm-5.3Input: $1.4 to $0.2219; Output: $4.4 to $3.39; Cache read: $0.26 to $0.1775
openrouter/z-ai/glm-5.3-flashInput: $0.045 to $0.15; Output: $0.6 to $0.5; Cache read: $0.0285 to $0.03
openrouter/~x-ai/grok-latestInput: $1.6 to $2; Output: $4.8 to $6; Cache read: $0.4 to $0.5
perplexity/openai/gpt-5.6-solInput: $5 to $4; Output: $30 to $20; Cache read: $0.5 to $0.4
us.meta.llama3-1-405b-instruct-v1:0Input: $5.32 to $2.4; Output: $16 to $2.4
us.meta.llama3-1-70b-instruct-v1:0Input: $0.99 to $0.72; Output: $0.99 to $0.72
us.meta.llama3-2-11b-instruct-v1:0Input: $0.35 to $0.16; Output: $0.35 to $0.16
us.meta.llama3-2-90b-instruct-v1:0Input: $2 to $0.72; Output: $2 to $0.72

The registry also updates capability flags, context/output limits, non-token rates, and deprecation dates on 244 more entries

Removed catalog entries (1)

azure_ai/muse-spark-1.3

New providers​

  • Add Prism provider - PR #41961
  • Add Sail as a provider, with service_tier mapped to its completion window - PR #42840
  • Add Cortecs as an OpenAI-compatible provider - PR #43872

Amazon Bedrock​

  • Surface a converse-stream 200 that decodes to no events as a 502 instead of an empty turn - PR #43213
  • Keep the provider status code on unprocessable image errors - PR #43416
  • Add xai grok-4.7 pricing and sync llama, mistral large 2407 and minimax m2.5 prices - PR #43623
  • Add bedrock_mantle rows for claude opus 5.5 and sonnet 5.5 - PR #43647
  • Add openai gpt-6.1-sol global and base rows - PR #43758
  • Add openai.gpt-6.1-sol us geo cris and Mantle rows - PR #43763
  • Add beta header for output config in message - PR #43778
  • Set gpt-6.1-sol max output tokens to 131072 - PR #43782
  • Keep applicable beta headers - PR #43829
  • Add beta header for thinking display updates - PR #43832
  • Add beta for mid-conversation tool changes - PR #43833
  • Accept Converse messages with no content key - PR #43936
  • Serve gpt-5.6+ chat completions natively by default, with chat_completions/ opt-in for gpt-oss and grok - PR #44307

Anthropic​

  • Drop thinking blocks with empty thinking text, not just missing signature - PR #38049
  • Stand default cache points down when extra_body hides a direct client mark - PR #43341
  • Forward the per-turn-control beta to Azure AI Foundry - PR #43415
  • Add claude-sonnet-5-5 model pricing - PR #43586
  • Correct Claude Sonnet 5.5 capabilities and provider keys - PR #43587
  • Forward the dangerous-tool-use beta to Azure AI Foundry - PR #43934
  • Keep thinking display updates beta - PR #43969

Fireworks AI​

  • Route and list the auto, auto-instant and firerouter routers - PR #43641

Gemini and Vertex AI​

  • Stop importing the vertexai SDK in partner-model completion - PR #42274
  • Keep legacy bucket_name in credential resolution and add GCS_BATCH_BUCKET_NAME env var - PR #42803
  • Make Gemma fake streams work with traced Responses - PR #43147
  • Forward seed to the Gemini API instead of rejecting it - PR #43197
  • Consider tools when validating context caching min tokens - PR #43319
  • Preserve proxy_server_request in completion adapter - PR #43536
  • Forward the per-turn-control beta for per-message output_config - PR #43558

OpenAI​

  • Exclude fine-tuned and custom gpt-5-chat aliases from gpt-5 reasoning path - PR #43185
  • Add openai gpt-6.1-sol from the pricing page - PR #43738
  • Add openai gpt-6-astra ultrafast tier prices from the pricing page - PR #43745

xAI​

  • Add native xAI batches and files support - PR #42812

Hosted vLLM​

  • Keep reasoning_content on replayed assistant messages - PR #43599

Model catalog and pricing​

  • Sync openrouter prices and add perceptron-mk1.5 - PR #43246
  • Add fireworks us-only deepseek v4.1 flash priority prices - PR #43247
  • Add typesafe/jev-router to the cost map - PR #43248
  • Add fireworks priority prices for muse glimmer 30b and deepseek v4 flash vision exp - PR #43252
  • Correct fireworks_ai deepseek-v4p1-flash pricing - PR #43253
  • Registry audit 2026-09-26, MAI-Image-2.5-Flash price, Databricks Claude Opus 5.5, Azure Foundry retirement dates - PR #43254
  • Remove duplicate openrouter/perceptron/perceptron-mk1.5 entry - PR #43273
  • Price fireworks deepseek v4.1 flash at the prices api value - PR #43311
  • Sync OpenRouter, Together, Cohere and Azure AI registry values with official sources - PR #43337
  • Correct azure gpt-4o-mini tts, transcribe, alias and MAI-Image-2.5 prices - PR #43357
  • Sync openrouter prices for deepseek, minimax, qwen and glm rows - PR #43384
  • Drop stale cache hit field from openrouter deepseek-v4-pro-0813 - PR #43389
  • Update azure_ai/grok-4.6 input price from Azure pricing page - PR #43440
  • Add azure_ai/MAI-Cyber-1-Flash - PR #43446
  • Sync openrouter prices from the models API - PR #43506
  • Add deprecation_date to together_ai Salesforce/Llama-Rank-V1 - PR #43507
  • Set together_ai gpt-oss-20b and gemma-4-31B-it deprecation_date to 2026-09-15 - PR #43509
  • Add mistral ocr pricing and azure max output limits - PR #43530
  • 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 - PR #43566
  • Add web search flag and model page source to anthropic claude-sonnet-5-5 - PR #43584
  • Add tool calling and reasoning flags, correct max output for nebius DeepSeek-V4.1-Flash - PR #43588
  • Take azure limits for deepseek-v4-flash-0731 and v3.2-speciale - PR #43597
  • Align Azure, Bedrock, Copilot, Gemini, Groq, OpenAI and OpenRouter entries with official docs - PR #43598
  • Add Vertex batch cache prices to vertex_ai/claude-sonnet-5-5 - PR #43602
  • Add baseten DeepSeek-V4.1-Flash-Fast - PR #43735
  • Lower fireworks up-to-4b size tier to the pricing page price - PR #43740
  • Add azure and openrouter gpt-6.1-sol rows - PR #43744
  • Take azure context limits from models-sold-directly - PR #43759
  • Add deprecation_date to two together_ai nvidia rows - PR #43809
  • Add fireworks priority prices for ember-1, nemotron and glm 5.3 us rows - PR #43811
  • Add Gemini Veo, Mistral and Azure Claude 4.5 deprecation dates - PR #43857
  • Add openai gpt-image-2.5 batch prices from the pricing page - PR #43869
  • Add vertex_ai gemini-3.8 flash tts rows - PR #43876
  • Add deprecation date for anthropic claude-sonnet-4-5 - PR #43898
  • Add perplexity, openrouter, voyage and nebius models and fix registry metadata - PR #43907
  • Raise baseten DeepSeek-V4.1-Flash max output to 262144 - PR #43916
  • Add fireworks inkling priority prices from the prices api - PR #43949
  • Sync openrouter prices from the models API - PR #43950
  • Set supports_vision true on GLM-5.3-Flash - PR #43951
  • Reprice fireworks deepseek v4.1 flash to the 2026-10-01 pricing update - PR #44024
  • Add vertex_ai/xai/grok-4.7 pricing - PR #44059
  • Restore later azure Models API retirement dates and date gpt-6.1-sol - PR #44072
  • Sync openrouter prices from the models API - PR #44105
  • Add azure_ai deprecation dates from the Azure retired models page - PR #44142
  • Take azure_ai claude-sonnet-4-5 retirement date from the Azure schedule - PR #44145

LLM API Endpoints​

Responses API​

  • Stream guardrail pre-call block as SSE with a typed output item - PR #42507
  • Fall back on pre-output stream drops, fail truncated streams, honor request_timeout - PR #43133
  • Record streamed /v1/responses container ownership before the response.completed frame - PR #43140
  • Hold Responses lifecycle events until output so a pre-output fallback announces one response - PR #43238
  • Run stream failure and success hooks on the iterating loop instead of blocking it - PR #43270
  • Emit the reasoning item on streaming /v1/responses for signature-only thinking - PR #43414

Anthropic Messages API​

  • Send a real error event when a /v1/messages stream fails - PR #41826
  • Surface Responses bridge stream failures as Anthropic error events - PR #43126
  • Rename the litellm.llms.anthropic.experimental_pass_through package to pass_through - PR #43329
  • Stream /v1/messages lifecycle frames live when no fallback can take over - PR #43600

Agents and Agent-to-Agent​

  • Add identity storage and validation contracts - PR #43720
  • Enforce authoritative agent permissions - PR #43721
  • Authenticate Entra identities and delegated requests - PR #43722
  • Add identity registration and dashboard controls - PR #43723

litellm.agent() (SDK)​

  • Add litellm.agent() to run Claude Code, Codex, OpenCode and Deep Agents through the AI gateway - PR #43885

Vector Stores and RAG​

  • Enforce key/team vector_stores allowlist on /v1/rag/query - PR #43953

Audio​

  • Honor base_url alias for Groq Whisper and report it as the api base - PR #43917

Pass-through endpoints​

  • Strip caller credentials from websocket passthrough - PR #43855
  • Relay Azure passthrough body model groups through the router - PR #43896
  • Preserve decision request bodies under token limits - PR #43920

General​

  • Keep _litellm_* kwargs out of provider request bodies by construction - PR #43221
  • Validate stream_chunk_size once, before any provider call - PR #43222
  • Keep the submitted body out of 422 validation errors - PR #43231
  • Salvage concatenated JSON tool call arguments - PR #43260
  • Count Gemini function_declarations tools - PR #43417
  • Categorize internal param appropriately to prevent leaking into request - PR #43783
  • Keep silent_model out of embedding provider requests - PR #44064

Management Endpoints / UI​

Admin UI​

  • Right-align money and count columns across tables - PR #37889
  • Render access group MCP and agent selections as wrapping chips - PR #41228
  • Surface x-litellm-call-id in Logs search, table and drawer - PR #42436
  • Filter tags by name and description on the Tag Management page - PR #42949
  • Rename All Models tab to Deployed Models and model filters to All Proxy Models - PR #43638
  • Add model leaderboard page - PR #43649
  • Show the user who owns the key that discovered a tool - PR #43892
  • Adopt the new LiteLLM logo and monogram - PR #43913
  • Drop the Beta badge from the Cost Optimization nav item - PR #43967
  • Give model leaderboard a distinct trophy icon - PR #44036
  • Show daily token totals on the model leaderboard - PR #44044
  • Leave unset callback select params out of the save payload - PR #44213
  • Shrink the sidebar logo so it stops outweighing page titles - PR #44247

Usage and analytics​

  • Aggregated daily activity endpoints return the top 100 keys in breakdown.api_keys by default (totals unchanged); set api_key_limit up to 1000 - PR #43398
  • Bounded daily activity routes (aggregated, search, model_top_keys, export, cache_leakage_keys) for all usage entities - PR #43408
  • Usage pages consume bounded daily activity routes instead of storing all keys client-side - PR #43409
  • Recover session key owners from daily spend for usage attribution - PR #43642
  • Look up hashed key names with two spend log rows per key - PR #43656
  • Add native ROI calculator for gateway spend vs merged PRs - PR #43669
  • Attribute completed batch cost rows to /batches in daily activity - PR #43870
  • Keep NULL entity ids when excluding entity ids - PR #44139
  • Reject non-canonical daily activity dates - PR #44143

Keys, teams and authentication​

  • Delete large teams without per-member transaction fan-out - PR #42998
  • Log key owner identity on expired key auth failures - PR #43105
  • Gate /sso/debug routes behind ENABLE_SSO_DEBUG, off by default - PR #43150
  • Persist SSO display name as user_alias on login - PR #44065

Proxy configuration​

  • Add LITELLM_DISABLE_LAZY_ROUTES to register optional routers at startup - PR #43911

CLI and coding agents​

  • Reuse saved agent setup and add reconfigure - PR #43392

Deployment​

  • Keep wheel paths under Windows MAX_PATH for Store Python - PR #43903
  • Drop the no-op PROXY_EXTRAS_SOURCE switch from the non-root image - PR #44097
  • Always exit when database setup fails at boot - PR #44141

Terraform​

  • Accept 2xx status codes in unified_access_group create - PR #42461

AI Integrations​

Lens and Agent Traces​

  • Add Rust storage foundation - PR #43819
  • Analyze agent activity with a separate worker - PR #43889
  • Agent traces tab on logs with timeline and otel setup guide - PR #43891
  • Correct ClickHouse rollup partitioning, dedupe keys, and retention changes - PR #43901
  • Port OTLP ingestion to current trace foundation - PR #43915
  • Store spend in ClickHouse automatically - PR #43928
  • Investigate sampled traces and retain batch results - PR #43942
  • Agent traces open in a side drawer with a chat-style run view - PR #43972
  • Improve trace ingestion and trace details - PR #43975
  • Track worker spend through virtual keys - PR #43989
  • Rename Lens internals and move its API from /engine to /lens - PR #44034
  • Inject tracing receiver and access context - PR #44035
  • Move traces and setup into Lens - PR #44068
  • Normalize agent spans in Rust - PR #44071
  • Add scoped SQL queries and schema-aware help - PR #44085
  • Simplify setup and investigation workflow - PR #44089
  • Add test trace, tracing key and otel endpoints to tracing setup - PR #44090
  • Label lens trace services as agents - PR #44116
  • Recalculate ClickHouse TTL info only on retention changes - PR #44117

Guardrails​

  • Honor experimental_use_latest_role_message_only on every request shape - PR #42447
  • Enable explicit PANW MCP output scanning - PR #43109
  • Honor litellm_params.timeout in every HTTP guardrail - PR #43134
  • Fix agent 365 to the production endpoint and log the opt-in fail_open at error level - PR #43189
  • Scan retrieved vector store chunks with the request's pre-call guardrails - PR #43271
  • Block private destinations in custom code http_request and bound guardrail execution time - PR #43280
  • Preserve Presidio output selection and restoration - PR #43401
  • Scan and mask top-level instructions with guardrails - PR #43629
  • Send a configured gateway_name from noma_v2 to Noma - PR #43678
  • Send request conversation and tool calls to post-call monitor - PR #43770
  • Scan Responses API input in Azure Prompt Shield - PR #43786
  • Treat an unknown straiker api_version as unset instead of skipping the guardrail - PR #43956
  • Scan Responses API input in Azure Text Moderation - PR #43965
  • Route Straiker sk_agt_ keys to v3 and fail closed on a missing verdict - PR #44011
  • Restore Azure guardrail get_user_prompt dispatch and allow logging - PR #44067

Logging and observability​

  • Add Databricks Zerobus trace logging callback - PR #42013
  • Keep the DataLakeServiceClient alive until its TTL elapses - PR #43082
  • Scrub PII and secrets inside object reprs and nested locals, add SENTRY_SEND_DEFAULT_PII opt-in - PR #43123
  • Keep team callback credentials out of the stored request body - PR #43217
  • Redact raw_request when turn_off_message_logging is set in the proxy config - PR #43219
  • Detach post-response service spans by request phase, name redis spans by operation - PR #43237
  • Excluded_services opt-out for datastore spans on tenant destinations - PR #43278
  • Add SigNoz preset for OpenTelemetry v2 - PR #43296
  • Deliver spans to app.langtrace.ai/api/trace with x-api-key - PR #43322
  • Send cache and reasoning tokens in langfuse usage_details - PR #43553
  • Keep request-body credentials out of stored spend-log requests - PR #43635
  • Add s3_partition_granularity option for hourly S3 folders - PR #43748
  • Register a UI-configured arize callback next to otel under OTel v2 - PR #43906
  • Name Data Lake objects without base64 padding or slashes - PR #43914
  • Keep tool payloads and logprobs unmasked in stored spend logs - PR #44075
  • Tolerate non-dict callback_settings.otel and ignore bare EXCLUDED_SERVICES env - PR #44086
  • Keep client call ids from sharing one Data Lake file - PR #44099

Spend Tracking, Budgets and Rate Limiting​

Cost tracking​

  • Keep the served service_tier on streamed chunks and spend rows - PR #42870
  • Record in spend logs whether a request used a client-forwarded Anthropic OAuth token - PR #43063
  • Bill chat per-second pricing once with a new cost_per_second field - PR #43614
  • Stop copying optional_params into response hidden params - PR #43637
  • Bill ultrafast prompts above 272k at the ultrafast long-context rates - PR #43764
  • Bill service tiers at catalog rates for custom-priced deployments - PR #43890

Budgets​

  • Count provider budget spend on every API surface - PR #38172
  • Email alerts at configured percentages of a team member budget - PR #42665
  • Resolve model_group_alias in the zero-cost budget predicate - PR #43512

Rate limiting​

  • Add fail_closed_rate_limit_enforcement to reject requests with 503 while Redis rate limit counters are unreachable - PR #43251

MCP Gateway​

  • Share compatibility-aware result conversion across tool surfaces - PR #43089
  • Add Microsoft 365 (Graph) server to the MCP catalog - PR #43099
  • Configure protocol versions and capability discovery - PR #43169
  • Report reachability without stored credentials - PR #43240
  • Align hub publication status and controls - PR #43241
  • Adopt shared server resolution and caller authorization - PR #43263
  • Scan and pin upstream tool descriptions - PR #43283
  • Scope OpenAPI listings to the exact server prefix and drop upstream OAuth metadata when a server is saved - PR #43608
  • Keep MCP permissions visible after key, team and MCP server saves - PR #43810
  • Resolve team-granted toolsets for non-admin keys and dashboard sessions - PR #43908

Performance / Loadbalancing / Reliability improvements​

Auto Router and model routing​

  • Parse the classifier verdict out of surrounding prose instead of falling to the default tier - PR #43215
  • Opt in to prompt-cache cost routing - PR #43232
  • Fetch cooldown state and usage counters in one Redis round trip - PR #43320
  • Compare historical and new savings consistently - PR #43348
  • Bind Claude Code background sessions to their auto-router - PR #43767
  • Strip encrypted reasoning the pinned deployment cannot decrypt - PR #43781
  • Carry per-request routing reads on context variables instead of public method kwargs - PR #43814
  • Honour the cooldown read interval in the routing prefetch - PR #43815
  • Show actual and baseline spend for historical savings - PR #44057

Caching, database and runtime​

  • Add maximum_daily_tag_spend_retention_period cleanup setting - PR #39221
  • Give SpendLogToolIndex its own share of the cleanup budget and log a per-run summary - PR #41768
  • Propagate auth cache invalidation over Redis Cluster via a node-level pub/sub client - PR #43110
  • Hold one spend counter batch across admission and across post-call accounting - PR #43369
  • One request-scoped Redis pipeline for auth, spend, rate-limit and routing reads - PR #43407
  • Run aresponses through the async wrapper so the cache is read once - PR #43769
  • Refresh auth management objects through the request Redis pipeline - PR #43776
  • One post-call Redis pipeline per backend for spend, rate-limit, routing and response-cache writes - PR #43779
  • Build the SpendLogs indexes in the migration job instead of in migrations - PR #43948
  • Write the response-cache SET to Redis at once instead of on the post-call batch - PR #43973
  • Gzip buffered responses for clients that accept it - PR #44052
  • Bound the lock waits of the partitioned SpendLogs index build - PR #44109
  • Hand libpq a root cert, not Prisma's sslcert, when the migration job builds indexes - PR #44203

Dependency updates​

  • Bump pyjwt, moment and brace-expansion to clear osv-scan - PR #43792
  • Bump oauthlib to 4.0.0 to clear osv-scan - PR #43899
  • Bump gitpython and tornado, extend diskcache osv ignore to Nov 1 - PR #43961
  • Bump pypdf from 6.16.2 to 6.19.0 - PR #44033

Documentation Updates​

  • Point readers to the security announcements mailing list signup - PR #43713

Tests, CI and Internal Changes​

These 98 PRs change tests, CI, contributor tooling, release packaging, or Rust runtime scaffolding that is not yet wired into a user-facing path. They do not change proxy or SDK behavior on their own

Tests (48)
  • Typed per-test metadata for the e2e suite - PR #42044
  • Record each e2e test's steps, starting with ProxyClient - PR #42393
  • Resolve the blank-S3 gateway repo root from the litellm package location - PR #42911
  • Pin async 5xx retry through the production AsyncHTTPHandler - PR #43080
  • Move tests into active CI selection - PR #43235
  • Pin the team-admin status-code matrix across every management route - PR #43249
  • Pin server resolution and authorization behavior - PR #43261
  • Classify every credential-bearing param for the canary suite - PR #43298
  • Credential canary suite harness - PR #43300
  • Sweep proxy logs, metrics, a Datadog intake and the Logs drawer for credential canaries - PR #43306
  • Request-path credential canary slots D1-D4 - PR #43307
  • Credential canary slots for MCP and pass-through credentials - PR #43308
  • Stored-config credential canary slots - PR #43309
  • Drain the logging worker after each logging callback test so no later test inherits its events - PR #43344
  • Stop VCR recording and replaying a test's own localhost upstream - PR #43346
  • Group /v1/messages contracts under tests/integration/messages_endpoint - PR #43352
  • Remove substring guard test_default_api_base - PR #43355
  • Native /v1/messages reasoning integration tests built on a captured Claude Code request - PR #43361
  • Read management routes back from the control plane replicas - PR #43373
  • Assert the usage the Gemma responses stream actually reports - PR #43422
  • Enforce shared upstream error contract in wheel checks - PR #43520
  • Pin org-admin status codes in the team-admin matrix - PR #43592
  • Callback credential canary slots C1-C3 and D5 - PR #43630
  • Refresh retired OpenAI tool-call models - PR #43676
  • Accept regional aliases that inherit Converse routing - PR #43785
  • Repair MCP Responses and budget fixtures - PR #43788
  • Settle the shared logging worker before recording shadow callbacks - PR #43847
  • Repair completion, SAIL, and spend-log fixtures - PR #43902
  • Restore the AWS env after a failed live call in the auth tests - PR #43921
  • Refresh qualified retired OpenAI fixtures - PR #43938
  • Scope user_api_key_auth overrides in proxy_server tests - PR #43952
  • Repair stale tests and move retired OpenAI text-completion fixtures - PR #43958
  • Repair stale tests and flaky CI infrastructure - PR #43983
  • Migrate DB and Redis backed proxy tests into tests/integration - PR #43996
  • Move auth, hooks, policy_engine and client tests into tests/unit/proxy - PR #43998
  • Move management_endpoints, management_helpers and guardrails tests into tests/unit/proxy - PR #44003
  • Move utils, agent_endpoints and endpoint tests into tests/unit/proxy - PR #44006
  • Inject the HIBP client and the MCP loop clock so two backend tests stop flaking - PR #44007
  • Move proxy_server, _experimental and db tests into tests/unit/proxy - PR #44012
  • Move middleware, spend_tracking, pass_through, common_utils and root proxy tests into tests/unit/proxy - PR #44015
  • Delete the legacy proxy test tree and shard tests/unit/proxy by glob - PR #44018
  • Bill Sail windows that synchronous calls can still use - PR #44058
  • Run the db push timeout hint test without a database URL - PR #44073
  • Keep 1ms-timeout deployments off the provider cache - PR #44082
  • Move live-provider legacy tests into tests/e2e - PR #44120
  • Move legacy proxy, router and Redis tests into tests/integration - PR #44128
  • Assert a saved Straiker api_version v1 with an sk_agt_ key routes to v3 - PR #44153
  • Repair stale and polluting tests red on scheduled main CI - PR #44229
CI (3)
  • Fail on new unbounded SQL IN lists and add a Prisma chunking helper - PR #42629
  • Sync the weekly release cycle with Linear releases - PR #43636
  • Test Redis behavior against local Redis and print short tracebacks - PR #44062
Code quality and contributor tooling (19)
  • Daily fresh tech debt cleanup, rolling PR (2026-09-25) - PR #43151
  • Add shared server resolver without changing callers - PR #43262
  • Replace Any with proven types in 5 files - PR #43304
  • Add the backport-stable label only for a P0 regression - PR #43351
  • Answer every team access check with TeamAccess.allows - PR #43364
  • Consolidate hub publication predicate - PR #43394
  • Clean up fresh tech debt from 2026-09-27 - PR #43538
  • Replace Any with proven types in 8 files - PR #43551
  • Drop UI, migration, and CODEOWNERS self owners - PR #43653
  • Clean up fresh tech debt from 2026-09-28 - PR #43674
  • Replace Any with proven types in 7 files - PR #43704
  • Clean up fresh tech debt from 2026-09-29 - PR #43830
  • Replace Any with proven types in 7 files - PR #43844
  • Remove the LIT002 mutable-construction rule - PR #43971
  • Clean up fresh tech debt from 2026-09-30 - PR #43993
  • Point mcp_server test references at tests/unit/proxy - PR #44055
  • Drop unused pytest-postgresql dev dependency - PR #44056
  • Split the KeyActivityPanel condition chains to bring the lint budget back under its ceiling - PR #44114
  • Remove banner comments, restating comments and dead in_loop_thread - PR #44161
Rust runtime internals (22)
  • Add the openai_like chat config foundation - PR #43379
  • Share anthropic types, request helpers, and streaming contracts across crates - PR #43426
  • Expand gateway configuration parsing - PR #43460
  • Share call lifecycle across route-owned inference - PR #43461
  • Add the HTTP host driver - PR #43462
  • Use shared execution in gateway inference - PR #43463
  • Support the HTTP Responses API - PR #43464
  • Connect Python inference bindings to shared routes - PR #43465
  • Add structured route lifecycle tracing - PR #43466
  • Separate gateway authentication and authorization - PR #43467
  • Add virtual key storage contracts - PR #43468
  • Add gateway UI login and sessions - PR #43469
  • Add the MCP gateway - PR #43470
  • Package the gateway container - PR #43471
  • Add litellm-db and litellm-db-testing workspace scaffolding - PR #43504
  • Remove delivery routing abstraction - PR #43514
  • Centralize host execution and compose callbacks - PR #43515
  • Select Rust caching through explicit cache objects - PR #43601
  • Orchestrate Messages route execution - PR #43719
  • Add shared llms wire type derives - PR #43730
  • Centralize Python bridge execution wrappers - PR #43871
  • Embed migration folders with a shared migrate! macro - PR #44104
Reverts of unreleased changes (2)
  • Revert the server-side Team Usage export beyond the top-N key cap, which never shipped in a stable release - PR #43376
  • Revert team key search beyond the top-N in the Team usage view, which never shipped in a stable release - PR #43377
Release and packaging (4)
  • 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 - PR #43789
  • Bump litellm-enterprise 0.1.72 -> 0.1.73, litellm-proxy-extras 0.4.103 -> 0.4.104 - PR #44126
  • Bump litellm-proxy-extras 0.4.104 -> 0.4.105 - PR #44235
  • Rebuild the Admin UI bundle on rc/1.105.0 - PR #44386

PR roll-up by ownership area​

Customer-facing PRs in rc.1: 238. Tests, CI and internal PRs: 98. Total: 336

  • Models & Providers: 81
  • Logging & Tracing: 36
  • LLM API Endpoints: 27
  • Performance: 26
  • Auth & Management: 19
  • Guardrails: 15
  • UI: 13
  • MCP: 10
  • Spend / Budgets / Rate Limits: 10
  • Docs: 1

New Contributors​

Full Changelog​

https://github.com/BerriAI/litellm/compare/v1.104.0-rc.1...v1.105.0-rc.1

LiteLLM Enterprise
SSO/SAML, audit logs, spend tracking, multi-team management, and guardrails, built for production.
Learn more →