Fireworks

Inference platform for open AI models

Fireworks is a platform for specialised intelligence, letting companies build, train and serve AI models fitted to their own data, workflows and products rather than adapting to a general-purpose model. It runs hundreds of state-of-the-art open models in production across text, image, embedding, audio and multimodal workloads. The engineering problems are infrastructure ones: low-latency inference and model serving that stays economical at scale. The company was founded by the team behind PyTorch. That lineage shows in where the company sits: the people who built the framework these models are written in now run the layer that serves them in production, which is where latency and cost are actually decided.