Choose the Effort per Chat, and See What Each Answer Cost

October 1, 2026 · InfraScout Team

You can now decide how hard the model thinks for each conversation, from a quick lookup to a deep investigation. Every answer also tells you which model gave it, at which effort, and what it cost, including the work it handed to Task Agents. Together, the two let you match the spend to the question and check afterwards that you did.

Pick the effort when you start a chat

The new-chat form has an effort control with four choices: Auto, Low, Medium and High. Auto is the behavior you had before, the model's own default. That is Medium on Claude Sonnet 5.5 and GPT-6, and High on most other models.

As a rule of thumb, Low fits lookups, inventory questions and tightly scoped playbook steps, where the answer comes from reading tool output rather than reasoning about it. High fits the hard investigations: a sign-in pattern that doesn't add up, a misbehaving server with three plausible causes, an attack path you want challenged. For everything in between, Auto is a sound choice.

Extra high and Max are deliberately not on the list. On the top models they think much longer on every turn, which makes them a cost decision for whoever runs the AI connection, not a per-chat toggle.

Some models don't support an effort setting at all, such as Claude Haiku 4.5. For those the control stays visible but disabled, with a note that it is not available for that model. Choosing a Model covers how the models and their defaults compare.

Change it between messages

If a conversation turns out harder, or easier, than you expected, click the effort badge in the chat header and pick a new level. The next message runs at the new setting. The badge is disabled while an answer is streaming or resuming, and it disappears once the chat has ended.

Set it up front when you can

Changing effort mid-chat means your next message re-writes the conversation's cached history once. On a short chat that is negligible. On a long one it is a one-off cost of roughly tens of cents. You only pay it on the message right after the change, so it is no reason to avoid switching. Just pick the right level at the start when you already know what kind of question you're asking. The prompt cache countdown explains the same cache from the other side.

See what each answer cost

The chat header has always shown what the whole conversation cost. Each answer now carries its own usage line underneath, for example:

Claude Opus 5.5 · Medium effort · 12k tokens · $0.0400

That tells you which answer was the expensive one, which model gave it after a connection switched providers, and whether the effort you picked made a difference. The line fades in when you hover or focus an answer, so a long conversation doesn't fill up with figures. On phones and tablets, a small Show turn usage toggle pins it open instead. The answer that is still streaming shows no line until it finishes. Answers from before this release show their model, tokens and cost, but no effort level, since it wasn't recorded then.

Delegated work counts toward the turn that delegated it

When an answer hands work to Task Agents, most of the spend can happen in those runs rather than in the answer itself. The usage line now shows both parts, such as $0.1200 + $0.4000 delegated, and adds · 2 running while runs are still in progress. Hover it for the number of runs and the turn's total. The turn that delegated the work no longer looks cheaper than it was, and the per-answer figures now line up with the total in the chat header.

Chats with many delegated tasks also open faster. A Task Agent card now loads its transcript only when you open it.

For admins: audits that look like the chat

The Chat Audit view now shows a conversation the way its user saw it, with the same layout, Task Agent cards and tool steps, instead of a separate timeline of message cards. The usage line is always visible there, so you can read through who spent what, on which model and at which effort, without hovering. Workflow Runs and Active Runs open the same view.

Get started

Start a new chat and set the effort before your first message. The Chat guide covers the rest of the chat experience, and Choosing a Model helps you match the model and its default effort to the work.

Questions or feedback? Reach us at info@infrascout.cloud.