Terms of exchange
Now to the unromantic matter of money. You load prepaid credit in advance, we charge per token as you spend it, and that is the whole of it. Your balance does not expire, and we do not enforce any hard rate limits. All figures below are per million tokens.
Highlights
Alibaba
Qwen 3.8 27B
256K token context
Input$0.80 /1M
Cached input$0.30 /1M
Output$6.00 /1M
Google
Gemma 4 31B
256K token context
Input$0.25 /1M
Cached input$0.06 /1M
Output$1.00 /1M
Moonshot AI
Kimi K3
256K token context, vision
PricingComing soon!
A complete account
| Model | Input /1M | Cached input /1M | Output /1M | Capabilities | Quantisation | Speed (median context) |
|---|---|---|---|---|---|---|
| Qwen 3.8 27B | $0.80 | $0.30 | $6.00 | 256K context coding |
FP8 (W8A16) | ~ 85 tok/s |
| Gemma 4 31B | $0.25 | $0.06 | $1.00 | 256K context roleplay, writing |
NVFP4 (W4A4) | ~ 50 tok/s |
| Kimi K3 | Coming soon! | - | - | 256K context coding vision |
NVFP4 (W4A8) | - |
| GLM 5.3 | Coming soon! | - | - | 1M context coding long-horizon tasks |
FP8 (W8A16) | - |
What the price procures
- No tokens squandered upon refusals, apologies, or lectures
- No prompt gymnastics to coax an answer from a direct question
- Pinned checkpoints, vetted and regression-tested against the originals
- One OpenAI-compatible API across every model we serve
- No content logs. We keep token counts and timestamps, nothing else
Terms of account
- Prepaid, pay as you go. Top up your balance in advance. No subscription, no seat count, no minimum monthly spend.
- Your balance never expires. Load what you wish, spend it when you wish.
- Volume discounts for substantial monthly use. Write to us as your volumes grow and we shall fix a rate worthy of them.
- Unused credit refunded on request. Your money remains yours until the moment you spend it.
- Best-in-effort technical support. Write to us with a difficulty and a person who built the thing will endeavour to answer it, fates be willing.
Enrol upon the waiting list
Demand has quite outstripped our machines. New accounts are admitted in batches as fresh hardware arrives. Leave an email address and we shall write to you the moment your slot opens.