Your model.
Our API.
Upload weights, pass rights review, and issue an OpenAI-compatible key. Billing, limits, and payouts stay on the platform.
Organization and model rights review · payouts from $20 · 25% markup
Licence
Apache-2.0
Author price
$0.15 / $0.45
Publishing
Weights uploaded
safetensors · 13.4 GB
Licence confirmed
Apache-2.0 · commercial use
Test run
A10G · p95 2.1s
Rights review
in the review queue
Publish and issue keys
after approval
From file to a paid endpoint.
Publishing is a sequence of checks, not a file upload into a shared catalog. Only after approval does a version get a public slug and start serving keys.
- 01
Organization
Legal entity, country, bank details, and a representative. KYB in the console.
- 02
Model and version
Weights, licence, architecture, chat template, and unit price.
- 03
Test run
Generation on the target GPU class before review.
- 04
Rights review
Rights to weights and data. No auto_map or pickle-only weights.
- 05
Keys and payouts
OpenAI endpoint, scopes, limits, ledger, and withdrawal requests.
Economics without surprises.
You set the unit price. Users pay it plus the platform markup; you receive the price you set. Every call becomes a separate ledger line.
- Payouts from $20
- 24 h cooldown
- Request status in the console
Your monthly revenue
$112.00
- User pays
- $140.00
- Platform markup (25%)
- $28.00
- To you
- $112.00
Requests × (input × input price + output × output price) ÷ 1,000,000. 25% markup is the default and can be changed by an admin.
A review you can pass.
Two independent layers: the organization and the model. A rejection states the reason; you can fix the version and resubmit it.
Organization · KYB
- Legal name and registration number
- Country of registration and address
- Representative and contact
- Registration document or extract
Model · rights and launch
- Rights to weights and training data
- Licence and commercial-use terms
- Supported architecture and chat template
- No auto_map and trust_remote_code
- No pickle-only weights (.pt, .bin, .h5)
- Test generation on the target GPU class
The list is not exhaustive: review may request more documents. Authors own answer quality; the platform checks safety and policy compliance.
A lab before production.
One prompt set across two versions: quality, p50/p95, error rate, and request cost. The baseline stays fixed so an update cannot silently regress the model.
| Set | v1.1 | v1.2 | Δ |
|---|---|---|---|
| support-en | 92% | 94% | +2 |
| json-mode | 98% | 99% | +1 |
| long-context | p95 2.4s | p95 2.1s | −0.3s |
Interface demonstration. Values are not production results.
AI labs
Publish checkpoints and update versions without running your own API gateway.
Proprietary models
Give partners key-based access with limits, without handing over weights.
Research teams
Compare versions on real requests and publish the stable one.
Author questions
Which models can I publish?
Models whose weights and commercial distribution rights you hold: LLMs, embeddings, images, ASR, and TTS. The architecture and weight format must be supported by the platform.
What does publishing cost?
Publishing and review are free. The platform takes a markup on actual calls; the author price stays yours.
How does review work?
Organization KYB first, then model rights and a test run. Status and notes appear in the console; a rejection always includes a reason.
When do payouts arrive?
A request is available from $20 and the amount is frozen until an admin confirms it. A 24-hour cooldown applies between requests. The bank transfer time depends on your bank.
Who owns the model?
The author. The platform does not change the set price retroactively or hand the weights to users.
Publish your first version.
Account, organization, and review request in one console.