Uncensored models. Hourly rates.
See Qwen 27B specifications, a measured decode result and its hourly rate. Each enabled session uses a dedicated GPU instance with managed model setup.
Funded-session model rates
| Model | Quantization format | Bit depth | Context window | Observed output tokens/s (one active stream) | Per hour |
|---|---|---|---|---|---|
| Qwen 27B | NVFP4 | 4-bit | 131,072 tokens (128K) | 104.98 measured | $2.19 |
| Qwen Flash-Next | Not available yet | ||||
| GLM Flash | Not available yet | ||||
Invite-only preview: credit is added manually. Checkout, deposits and bonuses are not offered. Sessions require operator enablement.
Observed Qwen decode: 104.98 tokens/s on 29 October 2026, with a 1,000-token prompt and 565-token answer. This is one measurement, not a speed guarantee; speed varies with the request. One generation runs at a time.
One generation runs at a time per session. A parallel generation returns 429 GENERATION_BUSY; wait for the active request to finish before retrying.
Choose your funded duration.
Prepaid wallet credit is reserved for a chosen duration before Start: a minimum of 15 minutes and a maximum of 24 hours.
Only elapsed READY time is charged. End session or expiry stops API access and billing. Unused reserved credit returns to your wallet.
Manual credit during the preview.
During the invite-only preview, credit is added manually. Checkout, deposits and bonuses are not offered.
Sessions require operator enablement. Qwen Flash-Next and GLM Flash are not available yet.
Add time while READY.
Use Extend session to reserve more wallet credit and add funded time while the session is READY. The new deadline applies after confirmation; extensions are never automatic.
Hourly rates. Billed by the second.
Rates are shown per hour, but billing uses elapsed READY time rounded up to the next second, including idle time. Startup is not billed.
Startup usually takes about 5 minutes and can take longer. This wait is excluded from paid time. New sessions require at least 15 minutes of funded time.
No per-token charges.
There’s no per-token charge or token meter on your bill while your funded session runs.
Each model has its own context window and generation speed. Run one generation at a time. Your client should wait and retry a 429 GENERATION_BUSY response rather than send more parallel requests.
End your session when you’re done.
Use End session when you’re finished. Closing the browser does not end the session.
End session and funded expiry stop access and billing. Closing the browser does neither; the funded deadline still applies.
Unused reserved credit returns to your wallet when the session ends.
The dashboard shows the funded deadline and warns as it approaches. Save needed responses in your client; API access stops at expiry.
Credit for service failures.
If a session fails because of our service, billing stops at the recorded failure time and unused reserved credit returns to your wallet.
How sessions start and end