{"version":"2.1.281","anchor":"sdk-set-max-thinking-tokens-reports-when-the-host-refuses-pa","canonical_anchor":"sdk-set-max-thinking-tokens-reports-when-the-host-refuses-pa","heading":"SDK set_max_thinking_tokens reports when the host refuses part of the request","tier":null,"area":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.281\/e\/sdk-set-max-thinking-tokens-reports-when-the-host-refuses-pa","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.281","markdown":"### SDK set_max_thinking_tokens reports when the host refuses part of the request\n\nSDK `set_max_thinking_tokens` now returns an error when the host refuses part of the request\n\n**Unclear.** The finding does not say which parts of the request a host may refuse.\n\n**What**\n\n`set_max_thinking_tokens` is an SDK request that sets how much the model may spend on thinking before it answers. The SDK is the toolkit programs use to drive Claude Code. If the host (the environment Claude Code runs in) refuses part of that request, the program that sent it now gets an error back: \"set_max_thinking_tokens: host refused part of the request\".\n\n**Why**\n\nA program that sets the thinking budget now learns when its request was not fully applied, instead of assuming it all went through."}