Product update · AI models
Effort and Thinking settings for Claude on your own key
When a prompt, persona, tool or assistant runs Claude on your own Anthropic key, you can now choose how hard it thinks before answering.
· 3 min read
- Shipped
- Area
- AI models
- Available on
- Suite add-on
Newer Claude models think before they answer. That helps on hard questions, but on a simple rewrite or a quick classification it just adds time and tokens. Until now you had no say in it: every request ran at the model's default effort. Now you can set it yourself, per item, for prompts, personas, tools and assistants that run Claude on your own Anthropic key.
What's new
Two new settings sit in the model settings next to the model picker:
- Effort. How much the model thinks before it answers, from low to max. Lower effort is faster and uses fewer tokens on your Anthropic account; higher effort suits harder tasks. Leave it on Default to keep the model's own setting.
- Thinking. Adaptive lets the model decide when to think. Off skips thinking, on models that allow it.
Both are saved with the item, so every run of that prompt, persona, tool or assistant uses them.
How to use it
- Make sure your Anthropic key is added in Studio. See bring your own API key.
- Open the prompt, persona, tool or assistant and pick a Claude model from your own key.
- Click the model settings (the sliders icon) next to the model picker.
- Choose an Effort level and, if you want, set Thinking to Off.
- Run it a few times to compare speed and answer quality, then save.
Good to know
- Only for Claude on your own key. The settings appear when the item uses a Claude model through your Anthropic key. Claude Haiku 5.5, included with your plan, runs at a fixed setting and doesn't show them.
- Not for chatbots. Effort and Thinking apply to prompts, personas, tools and assistants. Your chatbot's replies aren't affected.
- What you can pick depends on the model. Studio only offers the options the chosen model accepts. Some models can't turn thinking off at all, and some can only turn it off at high effort or below. If an option is missing, the model doesn't support it.
- Leaving both on Default changes nothing. Items you don't touch run exactly as before.
- Prompts, personas, tools and assistants are part of the Suite add-on. Bring your own key works on every plan; see pricing.
For developers
Requests on the Anthropic system accept two optional configuration keys. effort takes low, medium, high, xhigh or max. thinking takes adaptive or off. A value the chosen model doesn't support, or an unknown value, returns a readable error instead of being passed to Anthropic. Requests without either key are unchanged.