Fireworks AI AI features and pricing (October 2026)

Fireworks AI runs open models for developers and enterprises, from serverless per-token APIs to dedicated GPU deployments, and offers supervised and reinforcement fine-tuning on the same platform.

The platform supports more than 100 models for text, vision, audio and images, with batch inference, embeddings, reranking and function calling. Training ranges from guided fine-tuning to custom reinforcement learning loops. Fireworks names Cursor, Vercel, Notion and Sourcegraph as customers. According to its documentation, prompts and outputs of open models are not logged or stored without opt-in, and the privacy policy places the service's servers in the United States.

• Runs open models behind one APIServerless models cover chat, vision, audio and embeddings with per-token billing and postpaid invoicing. Developers test several models quickly before committing to one.
• Trains task-specific modelsSupervised and reinforcement fine-tuning adapt open models to a company's own tasks and deploy them right away. Teams can replace a large general model with a smaller specialized one.
• Serves production traffic on dedicated GPUsOn-demand deployments run a model on reserved GPUs with autoscaling and short cold starts. This gives steadier latency for applications with constant traffic.

• Zero data retention by default for open models; logging only with opt-in, according to the documentation.
• Reinforcement fine-tuning is available for models with more than a trillion parameters.

Fireworks.ai, Inc. runs the servers for its service in the United States, according to its privacy policy.

Servers are located in the United States, so check the transfer basis for EU personal data; standard contractual clauses are available on request. Note that the Responses API stores conversations for 30 days by default. GPU deployments cost $8.00 or more per hour, so compare against serverless usage.

No reviews yet. Be the first to share how your team uses it.

Every review shows its verification level. We read each one before it goes live. How we check reviews