Simple per-token pricing.
You buy credits up front, allocate them to API keys, and every request draws down the key by its exact token count, billed to $0.000001. One key works with every model we serve. No subscriptions, no minimum commitment beyond the first top-up.
| Model | Input · cache miss | Input · cache hit | Output |
|---|---|---|---|
glm-5.3availablefp8 | $1.40 | $0.26 | $4.40 |
glm-5.3-fastavailablefp4 | $2.80 | $0.52 | $8.80 |
glm-5.2availablefp8 | $1.40 | $0.26 | $4.40 |
glm-5.3 and glm-5.2 are served in fp8 at the base rate; glm-5.3-fast serves GLM-5.3 in fp4 on a faster lane with priority scheduling, priced at twice the base rate. Pick the model per request with the model field ("model": "glm-5.3"), or ANTHROPIC_MODEL=glm-5.3 for Claude Code — details in the docs. Prices are USD per million tokens. A cache hit is an input token whose prefix was already processed in a recent request; it is billed at a fraction of a miss, so multi-turn conversations stay cheap. Reasoning tokens are billed as output.
How credits work
- Top up your wallet by card: whole US dollars, from $5 to $1,000 per transaction. You are charged exactly the amount you enter.
- Create a key and choose how much credit to put on it. The key can spend up to that amount across any model.
- Adjust any time. Add credit to a key, reclaim what it hasn't used, or delete it — unused credit returns to your wallet. A key can spend slightly past its credit when requests run in parallel; that difference is charged to your wallet when the key is deleted.
- Track spend live in the console: per-key usage updates every 30 seconds, with a daily breakdown by tokens and cost.
Credits are prepaid and non-refundable except as set out in our Refund Policy. Need higher limits, invoicing, or a private deployment? Email contact@supernova-labs.ai.