Fireworks

The product is a training and inference platform for open AI models that helps teams customize models with their own data and run them in production with managed throughput, latency, and cost control. It sells to developers, engineering and data teams, and IT teams in businesses building AI applications and coding assistants. It is delivered as API-based infrastructure with serverless, on-demand, and reserved inference options and training workflows ranging from guided runs to custom RL, positioned as a drop-in replacement for closed-model APIs that routes to the best open or closed model per task.

Key features

• Training open models on custom data
• Reinforcement learning training workflows
• Multi-LoRA fine-tuning and serving
• Serverless pay-per-token inference
• On-demand dedicated deployments
• Reserved capacity with guaranteed quotas
• OpenAI and Anthropic compatible API
• Routing to best open or closed model
• Model library with single-line access
• Elastic global RL inference scaling