Credits

How building is paid for

Every generation runs on a plan and uses credits. There's no free generation allowance: each build uses real AI capacity.

Plans add credits for 30 days: 20 on Starter, 100 on Pro, 1,250 on Max.

• You pay for what a reply actually uses, measured in tokens (the pieces of text the AI reads and writes) and charged to the hundredth of a credit.
• Budget models (Claude Haiku, GPT-6 Luna, Gemini 3.8 Flash, DeepSeek, GLM, Mistral) are pure pay-per-token: you pay exactly for the tokens used, at the rate on the pricing page. A small change usually costs a few hundredths of a credit.
• Premium models have a minimum per build, from 2 credits (Claude Sonnet, GPT-6.1 Sol, Gemini 4 Argon) to 10 (Claude Fable 5.1, GPT-6 Astra). Bigger builds use more.
• Chat messages have a minimum of 1 credit on premium models; on budget models they're pay-per-token like everything else.
• Changes are cheap. Once a project has code, the AI sends only the lines that change instead of rewriting the whole file, so most follow-up requests use a small part of what the first build did.
• Repeated input (the instructions and your current code while one build continues) is often read from the AI provider's cache and charged at a quarter of the price or less.

The full list is in Choosing a model.

Seeing your usage

• Under every AI reply you'll see the exact tokens it read and wrote, as reported by the AI provider, and what it cost, for example DeepSeek V4.1 Flash · ↑ 12.2k ↓ 926 tokens · 0.02 credits. Hover for how much input was read from the cache.
• Settings → Usage shows the last 30 days: tokens and credits per day, by model and by project.

When you're charged

• Only when a generation succeeds. If a model is busy or a build fails, nothing is taken.
• If you don't have enough credits for the model you picked, you'll be asked to renew or upgrade your plan, or to pick a cheaper model.

Earning credits

Invite friends from Share & Earn: you get 5 credits for every new user who signs up with your link.