Supported AI Providers
Onysoft AI Gateway brings model families from many different AI companies together behind a single OpenAI-compatible endpoint. The catalog currently holds 739+ active models spanning text and chat, image generation, video generation, music and audio, and multimodal categories. On your side there is one agreement, one API key and one invoice — no separate account, payment method or SDK for every brand.
What this buys you in practice is model independence. When you want a job handled by a flagship model today and by a far more economical model tomorrow, the only thing you change is the model field in your request; your architecture, authentication and billing flow stay as they are. It also means you are not locked into a single vendor: when pricing, quotas or licence terms change, the alternative families already sit behind the same key.
Diversity matters during incidents too. When one route to a model is blocked, the request is directed to another working route for that same model wherever possible; if a model is entirely unreachable at source, switching to an alternative in the same class is a one-line change. You can watch live component status on /status, and if you are unsure which model to pick, onysoft/auto (OnyRouter) can make the choice per request. The groups below summarise the catalog's providers by function; the current model list and pricing always live on /models.
Providers by Category
Flagship chat and reasoning models
The top quality tier of the catalog: closed-weight flagship models built for deep analysis, complex code generation, long-document processing and multi-step agent workloads. This group covers the Anthropic (Claude), OpenAI (GPT), Google (Gemini), xAI (Grok) and Moonshot AI (Kimi) families. They differ in context window, reasoning behaviour and the price/quality balance they strike; the fastest way to see which fits your workload is to run the same prompt against several of them side by side in the Playground. Detail pages: /models/claude, /models/gpt, /models/gemini, /models/grok
Open-weight and open-source families
Open-weight families for scenarios driven by cost, customisation needs or vendor independence. Meta (Llama), Mistral AI, DeepSeek, Alibaba (Qwen) and Z.AI (GLM) form the backbone of this group, and the catalog also carries models from brands such as Microsoft, NVIDIA, Amazon, AI21 and Nous Research. As the open-weight side has closed the gap with closed flagships over the past two years, moving high-volume work to these families has become a realistic option for most teams. Detail pages: /models/llama, /models/mistral, /models/deepseek, /models/qwen
Search, RAG and source-citing models
A distinct category for teams building question answering over an internal knowledge base, or web-connected answers that cite their sources. Cohere's Command family is designed for enterprise RAG and search workflows, while Perplexity's Sonar models return answers alongside current web sources. In research, news monitoring and competitive tracking — where recency and verifiability are the point — these two families do a different job than general-purpose models. Detail pages: /models/cohere, /models/perplexity
Image generation and editing
Text-to-image and image-to-image editing models are reached through the same key as text models, via the asynchronous generation endpoint. This group covers Flux, Google (Imagen and Nano Banana), ByteDance (Seedream), OpenAI (GPT Image), Ideogram and xAI (Grok Imagine). Each family leads on something different: photorealism for some, rendering legible text inside an image for others, speed and unit cost for the rest. For product shots, ad variants and editorial illustration, running the same prompt through a few models is the most practical way to choose.
Video generation
Text-to-video and image-to-video models are listed as their own type and run asynchronously: the request returns a task id and you poll for the result. The group includes Google (Veo), Kling, Runway, Wan, MiniMax (Hailuo) and xAI (Grok Imagine). These models diverge sharply on resolution, clip length, motion consistency and cost per generation — and because video is priced per generation rather than per token, budgeting works differently than it does for text models.
Audio and music
Speech synthesis and music generation sit behind the same API key. ElevenLabs contributes multilingual text-to-speech (Turkish included), sound-effect generation and audio isolation models; Suno covers vocal or instrumental song generation, track extension and vocal separation. For call-centre announcements, voice assistants, video narration and content soundtracks, these families can be chained into the same pipeline as your text models.
Frequently Asked Questions
Which AI providers are supported?
The catalog holds 739+ active models from 40+ providers. On the text side the prominent families are Anthropic, OpenAI, Google, Meta, Mistral, DeepSeek, Alibaba (Qwen), xAI, Moonshot (Kimi), Z.AI (GLM), Cohere and Perplexity; on images, Flux, Imagen, Seedream, GPT Image, Ideogram and Grok Imagine; on video, Veo, Kling, Runway, Wan and Hailuo; on audio and music, ElevenLabs and Suno. The complete, current list is filterable by provider on the /models page.
How quickly is a newly released model added?
The catalog is refreshed on a regular automatic sync, and routine additions typically appear within a day. For notable launches the model is promoted manually and made available the same day. If you cannot find a model, verify the active list with GET /v1/models and open a support request if it is missing.
What happens if a provider has an outage?
When one route to a model is blocked, the request is automatically directed to another working route for that same model wherever possible. If a model is entirely unreachable at source, no infrastructure can compensate for that — but your advantage is that switching to an alternative in the same class means changing the model name in your request, nothing more. Live component status is published on /status.
Do I choose which provider is used?
Yes. You specify the model id in your request, and no silent substitution to a model you did not choose happens behind the scenes. If you would rather delegate the decision, put onysoft/auto in the model field to use OnyRouter: it analyses each request and picks a suitable model, routing costs nothing extra, and billing follows the selected model's own price.
Does pricing vary by provider?
Yes, and the spread between providers can be substantial. Text models are priced per million input and output tokens, while image, video and music models are priced per generation. All prices are converted to Turkish Lira at the current central-bank rate and deducted from your balance; see /models for current pricing and /calculator for a scenario-based estimate.
Do all providers use the same endpoint?
There are two. Text, chat and multimodal models run through the OpenAI-compatible POST /v1/chat/completions, synchronously or streaming. Image, video, audio and music models are asynchronous: start a job with POST /v1/video/generate and fetch the result with GET /v1/video/status/{task_id}. Both use the same sk-ony- key and the same balance.
Does switching providers mean changing my code?
No — moving between text models changes only the model name. Because the API is OpenAI-compatible, pointing your existing OpenAI SDK code at https://api.onysoft.com/v1 is enough; the request and response schemas stay the same. Switching across types (text to video, for example) only requires adapting to the generation endpoint's asynchronous flow.
Every provider behind one key
Create a free account, get your API key and start using the whole catalog from one endpoint.
Get Started Free Browse Models