Publish Your Model

List the model you trained or fine-tuned in the Onysoft catalog and let us handle billing, the API layer, documentation and customer support.

We review your application and run a technical evaluation. Once we confirm the model works we sign a Model Publisher Agreement with you and list it in the catalog. The whole process usually takes 5-10 business days.

What we provide

OpenAI-compatible API layer, key management and rate limiting
TRY billing, balance management and e-invoice infrastructure
Catalog page, documentation and Playground access
Server and GPU hosting support if you want it
Usage reporting and a customer support channel

Fields marked with an asterisk are required.

Applicant

Applicant type

* Legal name / Full name

* Contact person

* Email

* Phone

Website

Tax ID

Required for the agreement and payments.

Address

Model Details

* Model name

The name shown in the catalog.

Version

What does the model do?

What it is good at, what it was trained on, who it suits.

Is this a fine-tune of an existing model?

Base model

e.g. Llama-3.1-8B, Qwen2.5-7B, Mistral-7B

Base model licence

IMPORTANT: if the licence does not allow commercial use we cannot list the model.

Parameters (billions)

e.g. 7, 8, 70

Context window (tokens)

Input types

Text Vision Audio

Supported languages

Please state Turkish performance separately.

Quantization

fp16, int8, gguf-q4 etc.

Rights and Training Data

Training data source

What was it trained on? Does it include copyrighted or personal data? Be explicit — problems that surface later bind both of us.

Licence / model card link

* I declare that I have the right to publish this model commercially.

Tick if the model is your own work or its licence permits commercial publication.

Hosting

Where will the model run?

Endpoint URL (if self-hosted)

API format

Authentication method

Uptime you commit to

e.g. 99.5%

GPU required (if Onysoft hosts)

e.g. A100 80GB, L40S 48GB

VRAM required (GB)

Performance

Average throughput (tokens/second)

Time to first token (ms)

How many concurrent requests can it handle?

Measured on a single request producing a typical-length answer.

Benchmark results

MMLU, TruthfulQA, your own tests — share what you have.

Commercial Expectations

Expected input price (USD / 1M tokens)

Expected output price (USD / 1M tokens)

Revenue share you expect

This is your request; the final catalog price is agreed together.

Compliance

Does your model or infrastructure retain user prompts?

Critical under data-protection law. If it does, we state it plainly on the catalog page.

Content safety measures

Harmful-content filtering, system prompt constraints, red-teaming etc.

Anything else

Agreement

* I have read and accept the Model Publisher Agreement — I confirm the information I provided is accurate and that I have the right to publish this model.