Run times

How long a turn takes depends on what it actually has to do, not on the length of the prompt that started it. A short prompt that turns out to need a dozen tool calls will take longer than a long, detailed prompt that only needs one.

The biggest factor: tool call count

The biggest factor is the number of tool calls a turn ends up making.

Your own tools add real time

Running your own tools adds real time that has nothing to do with the model itself.

Thinking effort trades depth for speed

The thinking effort level (/effort low, medium, or high, see Thinking process) trades depth against speed directly: a higher effort level means the model spends more time reasoning through a decision before acting on it, which tends to reduce the number of wrong turns and follow-up corrections a task needs, at the cost of each individual step taking a little longer.

Why there is no fixed runtime estimate

None of this is fixed or predictable down to a specific number, and tCode does not report a runtime estimate up front, since it genuinely does not know how many tool calls a task will need until it starts working through it.

Waiting on a prompt does not count against runtime

Waiting for a permission prompt does not count against how long a turn takes in any meaningful sense, since tCode is simply paused, using no time or resources, until you answer.

Network conditions

Network conditions affect every remote call the same way they would for any other tool that talks to a server: a slow or unstable connection adds latency to each round trip with the model, independent of how complex the task itself is.

Connector references add a small, fixed cost

A turn that pulls in a connector reference, an email, a Notion page, a GitHub repository, a Linear issue, adds one resolution step at the point the message is sent, fetching the referenced content server-side before the model sees it.

Interrupting does not cancel a turn

Interrupting while a turn is running does not cancel it: the turn keeps going in the background exactly as it would have otherwise, only the local prompt stops waiting on it, so you get control of the terminal back immediately rather than being blocked until the turn finishes on its own.