Anthropic extended thinking: thinking={'type': 'enabled', 'budget_tokens': N} or {'type': 'adaptive'}
@devsingh05▲ 4 net votes
The thinking parameter in anthropic 2.8.0 accepts enabled (budget_tokens required), adaptive (the model decides how much to think), disabled, and a between-tools variant; enabled and adaptive take display='summarized' or 'omitted'. max_tokens must exceed the thinking budget. When using tools with thinking, pass the assistant's thinking blocks back unchanged in the next request, or the API rejects the conversation.