Together AI — AI Glossary

Reference details

Related terms

• Groq
• Fireworks AI
• OpenAI-Compatible API

How it works

1. Open-weight models are served behind an endpoint shaped like the common provider convention, so client code ports with a base-URL change.
2. Serving is optimised at the kernel and scheduler level, which is where providers in this category differentiate rather than on the models themselves.
3. Fine-tuning is offered against the same catalogue, usually with parameter-efficient methods, and the resulting adapter is served alongside the base model.
4. Capacity is available serverless per token or as dedicated instances, which is the lever for predictable latency under load.
5. Because the weights are open, the same checkpoint can be self-hosted later — the property that distinguishes this class from a proprietary provider.

Example: What open weights buy at the contract level

Two providers serve the same open model at similar prices. One is a hosted API for proprietary models; the other serves open weights. If either raises prices or deprecates the model, the open-weight path has an exit: download the checkpoint and self-host, at a known engineering cost. The proprietary path does not, so the same prompt must be re-tuned for a different model. The leverage is not in today's price — it is in what the alternative costs if the price changes.

Commonly confused with

One of several hosts serving open-weight models behind an OpenAI-shaped API, alongside Fireworks, Replicate and Groq. They differ less in the models offered than in what they optimise: dedicated capacity and fine-tuning here, versus latency-specialised hardware or per-container flexibility elsewhere. Since the API shape is shared, the switching cost is low and the evaluation should be on latency, price and model availability for your specific model.

When to use it

Reach for it when:

• Serving open-weight models without operating accelerators.
• Fine-tuning an open model and serving the adapter through the same provider.
• Keeping a credible path to self-hosting the same weights.

Reach for something else when:

• A frontier proprietary model is required.
• Data residency or private networking requirements a hosted API cannot meet.
• Sustained high volume where self-hosting is already cheaper.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Together AI — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path