Plans that grow with your research

Each plan comes with a monthly allowance, measured in credits, that covers AI chat, the research assistant and GPU training. Larger models, higher thinking effort and bigger GPUs use it faster. You can see what's left in your account at any time. On a paid plan, unused credits expire at the end of each month. On the free plan, AI usage comes in 10-hour windows instead, and what you don't use in a window expires when it ends.

You can top up credits at any time, or move to a plan with more usage. On the free plan you can also wait: your usage comes back when the next 10-hour window opens. A GPU run that's already going isn't stopped the moment credits run out: it gets a grace period of up to an hour to finish, then Ctrl+C so your code can save a checkpoint, and the machine stops.

Pick a tier, and a specific card in it if you like; every card in a tier costs the same. It's charged per minute, from the moment your GPU machine is ready. Start-up time and machines that fail to start are on us. A machine with nothing running stops after 15 minutes, with a warning first. Each plan has a longest job; you're warned 15 and 5 minutes before it ends. On the free plan, GPU sessions cost nothing, within the free limits.

On the free plan, AI usage comes in 10-hour windows. A window starts with your first message after the last one ended; what you don't use expires when it ends. There's also a weekly limit: once you reach it, the next window opens when the week resets, on Monday at 00:00 Vietnam time. Free GPU time is an hour a month on Starter cards, renewed on the 1st: use it in one session, or split it across shorter ones with 2 hours between them. Start-up time and machines that fail to start don't count. Paid plans have none of these waits.