Ask anything.
Nobody can read it.
Every model. A label that says who can read it. Prices below the labs.
Locked on your device. Opened only inside the chip.
Encrypted to one verified chip.
We move bytes we can't read.
The model runs inside the hardware.
The chip signs what ran.
Three ways in.
Every model tells you who can read it.
Private shouldn't cost extra.
Supply is unused credit, listed by the people who hold it. Output, per million tokens.
| Model | Who can read it | List | Here |
|---|---|---|---|
| Claude Fable 5.1 | anonymized | $50.00 | $10.00 |
| GPT-6 Astra | anonymized | $50.00 | $11.50 |
| Claude Sonnet 5 | anonymized | $10.00 | $2.20 |
| Kimi K3 | sealed | $13.30 | $3.90 |
| DeepSeek V4 Pro | sealed | $1.74 | $0.61 |
| Llama 4 Maverick | sealed | $0.85 | $0.32 |
| Qwen 3.8 235B | private | $1.20 | $0.38 |
| Gemini 3.8 Flash | anonymized | $3.75 | $0.75 |
- Deposit stays in dollarsFrom $5. Pay per request.
- Better prices serve firstSellers set 10% to 80% off.
- The real model, provablyEvery reply carries the weight hash.
Two lines to switch.
Change the base URL and the key. Everything else works as it does today.
- One key, every model
- Privacy floor per request
- Agents pay per call, no account
- Receipt on every response
Unused credit expires. Sell it instead.
List a key you don't fully use. Set your discount. Earn in USDC.
- Paste a keyPick how much to offer and at what discount.
- It's sealedNo employee can read it. No buyer ever sees it.
- Earn per requestWithdraw from $10. No payout fee.
Straight answers.
Is Claude really private here?
Partly. Anthropic sees the text, not you. We keep nothing. For nobody-can-read-it, pick a Sealed model.
What does "Sealed" mean?
Encrypted in your browser to a key only one verified chip holds. We route ciphertext we cannot read.
Where does the discount come from?
People sell credit they won't use. You pay the best offer. We take a small spread.
Do you store my prompts?
No. Model, tokens, cost and timing only. Never the text.