Models and limits
What is available today, how big a request may be, and how much you may use.
Models
| Model | API string | Alias | Per call | Typical speed |
|---|---|---|---|---|
| QED 1writer | qed-2026b | qed | 2,000 words in; voice passages 1,500 words; document context 1,200 words | 3 s for 200 words |
| Sic 1detector | sic-2026a | sic | 10,000 words | 2 s per 1,000 words |
| QED Minor | forthcoming Smaller and faster. | |||
| QED Major | forthcoming Largest. | |||
| Sic Minor | forthcoming Smaller and faster. | |||
| Sic Major | forthcoming Largest. | |||
Dated strings are fixed checkpoints. Aliases move to each family's current default when a new one is released. A replaced checkpoint stays callable by its dated string: qed-2026a, the first release of QED 1, was superseded by qed-2026b.
Both models are kept warm, so there is no start-up delay. Neither streams its output.
Limits
| Requests per minute | 20 per account by default. |
| Concurrent requests | 4 per account by default. |
| Monthly allowance | $5.00 of use at list price by default. It resets on the first of the month, UTC. |
| Request size | 2 MB. |
Limits are per account, not per key. If you need more for a study or a class, write to hello@etal-labs.com and say roughly how much.
Prices
There is no billing yet. Usage is counted against the monthly allowance at these indicative prices.
| QED 1 | $2.00 per million input tokens and $6.00 per million output tokens. A 200-word passage from a short brief costs about $0.002. Voice passages count as input on every call. |
| Sic 1 | $0.01 per 1,000 words. |
| Compare | Free without the judge. With it, the judge model's own cost, about $0.005 per check. |
QED checks each output before returning it and regenerates one that fails. Regenerated samples are metered, and the response says how many were taken.
What we keep
- Not your text. The text you send and the text the models return are processed in memory and not written to our database or logs.
- Request records. Endpoint, model, sizes in words and tokens, cost, latency and status, so that you can see your usage and we can enforce limits.
- Account. Your email address, and your name if you sign in with Google. API keys and sign-in tokens are stored as hashes.
- Processors. QED runs on Fireworks AI and Sic on Modal; your text passes through them to be processed. The privacy notice lists everyone involved.