Plans that grow
with the work.

Free includes Ken in chat, with no card and no time limit. It covers about 20 everyday messages a day, within a weekly limit, and the day refills at midnight Eastern. Paid plans add Sigel, thinking, Wynfall Code and much more usage.

Signup is open, and we let people in by hand, in waves, while internal testing finishes. Not everyone who signs up will be let in right away. Once you are in, you get 7 days of Builder, free and with no card: Sigel, extended thinking, Wynfall Code and Builder usage. The 7 days start when you are let in, not when you sign up. When they end, your account moves to Free on its own. Nothing is charged, and you keep Ken every day.

On paid plans, the models can think a problem through before they answer. Your plan sets how far they can go. Standard covers most everyday work. Extended gives more room for long documents and code that takes several steps. Max gives the most room, for the hardest problems. You choose the effort for each chat, and deeper thinking uses more of your usage. Free gives fast, direct answers without the thinking step.

For those who want the detail: we count usage in units. A token is a small piece of text, about three quarters of a word, and every token costs a set number of units. On Sigel, reading counts 1.5 units a token and writing 3.6. Ken counts less: 0.6 for reading and 2 for writing. Text already read earlier in a conversation counts a tenth (0.15 on Sigel, 0.06 on Ken), and thinking counts as writing. If a request processes 2,000 input tokens and generates 500 output tokens, that is 2,500 tokens of model work, or 4,800 units on Sigel. Repeated context and reasoning can add more, so a token count is not a message count. Our estimates assume a typical message reads about 1,500 tokens and writes about 1,700 including thinking (about 8,300 units on Sigel, or 2,300 on Ken without thinking), and that an hour of coding is about 20 steps of about 18,000 units each. Per request, Standard thinking can use up to about 8,000 tokens, Extended about 16,000 and Max about 32,000. Long conversations and deep thinking use more, short ones less. The weights and allowances are the ones on this page, and a meter shows what is left.

On Free, about 50 pages (32K tokens). On Starter, about 100 pages (64K). On Builder and Studio, about 200 pages (128K). A page is about 500 words. Rising as capacity comes online.

Start on Free. If you reach for Wynfall most days, Starter adds Sigel and Wynfall Code. Builder is the one we suggest for daily writing or coding, and Studio is for people who work with it all day.

One task is one request in flight, whether that is a chat message or a Wynfall Code job. Your plan sets how many can run at the same time, not how many you send in a day.

Paid plans get queue priority in the order shown in the table. Tasks at a time is a ceiling, not reserved GPU capacity, so busy periods can mean a short wait. We will publish queue, context and fair-use limits before subscriptions open.

When a paid plan runs out of usage for the week, Sigel pauses and Ken keeps working, at the lowest priority, from a separate share of the week. That share has a limit: it is there so you are never locked out, not as an extra allowance.

You never get locked out of a paid plan. When your usage runs out, Sigel pauses and Ken keeps working, under fair use, until your usage refills or you move up a plan. Move up at any time. We are also designing optional top-ups with a visible price and a usage meter. No automatic overage charges are planned, so a bill never grows without you choosing it.

Annual bills once a year at 17 percent off: $100, $200 or $500 instead of $120, $240 or $600. Your usage windows work the same as on a monthly plan. The price does not change during a year you have already paid for.

Yes, over time. The prices on this page are founding prices. As we add capacity and compute, prices will rise. If you subscribe at a founding price, you keep it for as long as you stay subscribed. If you cancel and subscribe again later, you pay the price at that time. Any price change, for anyone, comes with at least 30 days notice by email, never takes effect partway through a billing period, and you can cancel before it applies.

The API is a separate offering, priced by model. Rates, metering and spending controls will be published before API access opens. Consumer subscriptions do not include API credit.

No. Nova is still in research and is not part of any plan. There is no Nova launch date yet.

Sometimes. We plan to thank paid subscribers with occasional complimentary resets. They are a gift, not a guaranteed allocation, and any terms will be clear when one is offered.

Not yet. Sign up for early access: there is no card, and once you are let in you get 7 days of Builder. Paid plans open during early access, and early-access members get 30% off monthly plans while they stay subscribed. The offer does not apply to annual plans and ends October 22.