Together AI

What it does

Together AI runs a platform for building and running AI models. It offers serverless inference APIs, batch inference, provisioned throughput with reserved capacity, dedicated model inference on custom hardware, and dedicated container inference for video, audio, and image models. It also provides GPU clusters, fine-tuning, evaluations, custom training, code sandboxes, and managed storage. A model library covers text, image, video, code, and voice models. Documentation, pricing, a playground, and a sales contact are linked from the site. Sign-in is available with Google, GitHub, or SSO.

Serverless inference is described as running open-source models on demand with no infrastructure to manage. Batch inference is described as asynchronous processing that scales to 30 billion tokens per model. Provisioned throughput is described as committed capacity with token-based pricing and a 99% uptime SLA. Managed storage lists object storage and parallel filesystems with zero egress fees. GPU options named on the site include GB300, GB200, B200, H200, and H100. Models named include GLM-5.3, DeepSeek V4 Pro, Qwen3.8, Kimi K3, Llama 4 Maverick, and gpt-oss-120b. Together Link is described as running open models inside an existing harness. Published research listed includes FlashAttention, ATLAS, ThunderKittens, and DSGym. Developer resources include documentation, demo apps, and cookbooks. The contact sales page offers a platform walkthrough, an Enterprise trial, and a custom plan discussion.