Fast, Slow and Custom

Every conversation runs in a thinking mode. It decides how much the
assistant works a problem out in its head before it acts — and, if your
administrator has set it up that way, which model answers.

The chip sits at the left of the message input and always shows both values,
for example ⚡ Fast · Off. Click it, or press Cmd/Ctrl+Shift+M.

The two modes

	Mode	What it does	Good for
⚡	Fast	Answers from what it can see and look up, without stopping to reason it out first. The fastest setting.	Questions, lookups, "where is X", small edits, anything you will iterate on
◉	Slow	Works the problem out before it acts.	Design, subtle bugs, unfamiliar code, anything you would rather it got right first time

Both use the same model and the same tools. Fast is not a smaller
assistant — it is the same one, spending less time narrating to itself and more
time going and looking. On our own benchmarks it takes more steps than Slow,
not fewer, and finishes in about a third of the time.

The trade is real, though. On easy work the two finish equally often. On
hard work — the kind Slow is named for — Fast gets there markedly less often,
by a wide margin on our own hard-task benchmarks. So the rule of thumb is not
"Fast unless you need quality"; it is **Fast for the work in its column above,
Slow when the answer has to be right the first time.** Switching mid-thread is
one click, and a good habit: ask in Fast, and move to Slow when the answer
starts mattering.

If you are curious why the modes are named after a book — and what our
benchmarks actually say — see
Thinking, Fast and Slow.

Custom is the third entry: your own mix, chosen per role — Slow for the
work that matters, Fast for the background jobs like summarising and titling.
It appears whether or not you have ever set a per-role assignment, so trying a
mode and coming back is one click.

Your administrator may have replaced these with their own named set. If the
picker shows something different, the descriptions in it are theirs.

Deliberation

Inside each mode is one dial. It sets **how long the assistant reasons before
each step** — not how many steps it takes, and not which model answers.

Rung	What it does
Off	No deliberation — it acts, and looks up what it needs.
Minimal	A glance before it acts.
Low	A brief check before it acts.
Medium	The model's own habit, steered neither way.
High	Most of the thinking before the first move.
Extra high	The whole problem worked out before the first move.
Max	As long as it takes before the first move.

Each mode starts somewhere on this dial and offers only part of it: **Fast
starts at Off and covers the quiet end, Slow** starts near the top and
covers the deliberate end. Each model also supports only some rungs. A rung your model has
never heard of is not shown; one your model has but the mode does not offer is
greyed, with the reason in its tooltip. Hovering any rung you can pick tells
you what it does.

Step the dial without opening the popover with Cmd/Ctrl+Shift+↑ and
↓.

Switching, and what it changes

• Mode and dial belong to the conversation. Changing one never changes another, and never changes a colleague's.
• The change takes effect on your next message. Work already running finishes on the setting it started with.
• A marker appears in the transcript where you switched, so a conversation you come back to still explains itself.
• Background work — summarising the conversation, naming it, fetching a page — runs at the mode's own setting, not at whatever you have moved the dial to. Turning your conversation up to Max makes the assistant think harder; it does not make the summariser think harder too.

Use as default for new conversations at the bottom of the popover makes
your current mode the one new conversations start in. It leaves every existing
conversation alone.

From the keyboard

Type these in the message input:

• /mode slow — switch mode. Partial names work (/mode sl), and the retired names balanced and deep still land on Slow.
• /effort high — move the dial. /effort extra high and /effort x-high both work.
• /mode or /effort on their own — answer with what is currently in force.

Where else it lives

Settings → Extensions → Thinking Modes shows every mode as a card — its
description, the model it would answer with, the rung it starts at and a
relative cost badge — plus the role map: which model each background role
actually resolves to, and why. It is the answer to "so which model summarises
my conversation?" without reading any configuration.