LLM API migration errors and fixes

Every documented error and behavior change the request migration linter checks, each with its fix and the official source.

Anthropic (Claude)

Change	Status	Models
temperature, top_p and top_k return an error on Claude Sonnet 5.5	Rejected	`claude-sonnet-5-5`
Assistant message prefill returns a 400 on Claude Sonnet 5.5	Returns 400	`claude-sonnet-5-5`
tool_choice any or tool returns a 400 on Claude Sonnet 5.5	Returns 400	`claude-sonnet-5-5`
thinking type enabled with budget_tokens returns a 400 on Claude Sonnet 5.5	Returns 400	`claude-sonnet-5-5`
thinking type disabled returns a 400 on Claude Sonnet 5.5	Returns 400	`claude-sonnet-5-5`
Top-level output_format returns a 400 on Claude Sonnet 5.5	Returns 400	`claude-sonnet-5-5`
computer_20250124 isn’t accepted by Claude Sonnet 5.5 or Haiku 5.5	Returns 400	`claude-sonnet-5-5`, `claude-haiku-5-5`
fine-grained-tool-streaming-2025-05-14 with a toolset returns a 400	Returns 400	`claude-sonnet-5-5`, `claude-haiku-5-5`
Beta headers to remove when moving to Claude Sonnet 5.5	Behavior change	`claude-sonnet-5-5`
Claude Sonnet 5.5 thinks by default when thinking is omitted	Behavior change	`claude-sonnet-5-5`
thinking type enabled with budget_tokens returns a 400 on Claude Haiku 5.5	Returns 400	`claude-haiku-5-5`
Assistant message prefill returns a 400 on Claude Haiku 5.5	Returns 400	`claude-haiku-5-5`
temperature, top_p and top_k can return an error on Claude Haiku 5.5	Can return an error	`claude-haiku-5-5`
Forced tool_choice on Claude Haiku 5.5 skips thinking	Behavior change	`claude-haiku-5-5`
temperature, top_p and top_k return a 400 in API version 2026-10-21	Returns 400	`claude-*` (API 2026-10-21)
A document block in an assistant message returns a 400 in API version 2026-10-21	Returns 400	`claude-*` (API 2026-10-21)
citations on a user text block return a 400 in API version 2026-10-21	Returns 400	`claude-*` (API 2026-10-21)

OpenAI

Change	Status	Models
max_tokens isn’t compatible with OpenAI o-series models	Rejected	`o1`, `o3`, `o4*`
stop isn’t supported on o3 and o4-mini	Rejected	`o3`, `o3-`, `o4-mini`, `o4-mini-`
Tool calling on GPT-6 Astra, GPT-6.1 Sol and GPT-6.1 Terra requires the Responses API	Rejected	`gpt-6-astra`, `gpt-6.1-sol`, `gpt-6.1-terra`
Chat Completions function tools on GPT-6 Luna, GPT-6 Sol and GPT-5.6 need reasoning_effort none	Rejected	`gpt-6-luna`, `gpt-6-sol`, `gpt-5.6-*`
reasoning effort none returns a 400 on GPT-6 Astra	Returns 400	`gpt-6-astra`
GPT-6.1 Sol doesn’t support reasoning effort none or minimal	Rejected	`gpt-6.1-sol`
Remove temperature, top_p and logprobs on GPT-6 unless reasoning effort is none	Rejected	`gpt-6*`
temperature, top_p and logprobs raise an error on GPT-5 models	Rejected	`gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-5.2`, `gpt-5.4`
prompt_cache_retention becomes prompt_cache_options.ttl on GPT-6	Behavior change	`gpt-6*`

Last checked 2026-11-02 against the official OpenAI and Anthropic docs.