One model (Qwen3.8-27B), a dedicated server per customer, flat hourly, prepaid credits. Per-second billing. No egress fees. Zero data retention. Your server auto-stops after your configured idle time (default 15 min) or when credits run out (60s grace for in-flight generations). A new server takes up to 10 minutes to deploy.
| SKU | Capacity | Speed | Price |
|---|
$10 → 1,000 credits · $50 → 5,000 · $100 → 10,000 · $500 → 50,000.
1 credit = $0.01 of compute. Minimum top-up $10. Auto or manual recharge.
"Concurrent users" = concurrent in-flight generations (max-num-seqs), not unique humans. Micro uses 2-bit mixed precision (verbose, ~2× time — short tasks only). 1M context on Solo is YaRN-scaled from the native 262K (Qwen's recommended method).