Together AI was founded in 2022 by Vipul Ved Prakash and Ce Zhang, along with a team of researchers who had contributed to important work in distributed systems and machine learning. Based in San Francisco, the company has raised over $225 million in funding, including a $106 million Series A led by Kleiner Perkins.
The company’s core offering is a cloud platform optimized for running open-source AI models. While cloud providers like AWS and Google offer GPU instances, Together AI provides an inference and training platform specifically tuned for popular open models like Llama, Mistral, Stable Diffusion, and others. Their API is compatible with the OpenAI format, making it easy to switch.
Together AI has invested in research as well. They released RedPajama, one of the first open-source datasets for training LLMs, and have published work on efficient training techniques and model architectures. Their inference stack achieves high performance through custom optimizations for popular model architectures.
The platform offers three main services: inference (running models via API), fine-tuning (customizing models on your data), and dedicated GPU clusters for larger training runs. Pricing is competitive, often significantly cheaper than running equivalent workloads on major cloud providers. For developers and companies who want to use open-source AI models without managing their own GPU infrastructure, Together AI has emerged as one of the go-to platforms alongside options like Replicate and Fireworks AI.
