Together AI
A cloud platform for open models: serverless pay-per-token inference through an OpenAI-compatible API, dedicated single-tenant endpoints, fine-tuning on your own data, batch jobs and GPU clusters. Useful when you want open-weight models such as DeepSeek or Qwen without running your own GPUs.
Platform
AI & Machine Learning
Together AI
Under an hour to a first call
Intermediate
Pay per token; dedicated endpoints extra
What Together AI is
Together AI is a cloud platform built around open-source and open-weight models. It sits between calling a proprietary API and running your own GPUs: you get open models and their licences, and someone else runs the servers.
What it offers
- Serverless inference: call hosted open models through an OpenAI-compatible API, paying per token, with prices that depend on model size.
- Dedicated endpoints: single-tenant GPUs for predictable, isolated latency when traffic is steady or data must not share hardware.
- Fine-tuning: train a model on your own data and deploy it for inference on the same platform.
- Batch: queue large jobs asynchronously and collect the results later.
- GPU clusters: H100 and B200 clusters for training or very large jobs.
As of October 2026 its featured models include DeepSeek V4 Flash, Qwen3.8 and several other open model families for chat, vision, audio and images. The catalogue changes often; check the models page for exact names and context lengths.
Why it matters when choosing a stack
Open-weight hosts let you test a model's licence, quality and cost without buying hardware. Because the API is OpenAI-compatible, you can run the same golden test set against Together, another host and a proprietary API by changing only the base URL and model name. That makes it a practical option for the privacy and deployment choices in the Choosing an LLM roadmap, and a stepping stone towards self-hosting later.
Check the provider's data handling terms and regions before sending sensitive data, as you would with any API.
Prerequisites and time
An account, an API key and any OpenAI-compatible SDK. A first call takes minutes; a fine-tuning job needs a prepared dataset and more time.
Free or paid?
Paid per token for serverless use, with separate pricing for dedicated endpoints, fine-tuning and GPU clusters. See the pricing page for current rates.
Common questions
Serverless or dedicated? Serverless is cheapest for low or spiky traffic because you pay only for tokens. Dedicated endpoints make sense when traffic is steady and high, or you need isolated hardware.
Best for: teams that want open-weight models, fine-tuning or dedicated capacity without running GPU servers themselves.
Ready to start?
Opens on Together AI in a new tab.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.