What it does
Together AI presents itself as a full-stack AI platform built on its own research. Its home page groups the product into inference, compute and model shaping, and showcases featured open models for text, images, video, code and voice. On the inference side, serverless inference runs open-source models on demand with no infrastructure to manage and no long-term commitments. Batch inference handles large asynchronous workloads, scaling to 30 billion tokens per model. Provisioned throughput offers reserved capacity with token-based pricing and a 99% uptime SLA. Dedicated model inference deploys models on dedicated infrastructure for teams that want speed and control, while dedicated container inference is GPU infrastructure for generative media workloads, deploying video, audio and image models. On the compute side, accelerated compute ranges from clusters you can start yourself up to thousands of GPUs, with on-demand B200s mentioned. Code sandboxes give AI apps and agents somewhere to build and run code, and managed storage is described as having zero egress fees. Fine-tuning lets you adapt open-source models for production without managing training infrastructure.
Capabilities
- Serverless inference for open-source models
- Batch inference for large asynchronous jobs
- Provisioned throughput and dedicated deployments
- GPU clusters, from self-serve setups up to thousands of GPUs
- Fine-tuning of open-source models
- Code sandboxes and managed storage
When to use it
- You want to call open-source models over an API without running infrastructure.
- You need reserved or dedicated capacity for production inference.
- You need GPU clusters for training or large workloads.
- You want to fine-tune an open model without managing training infrastructure.
