Fireworks

Founded by Lin Qiao in 2022

Run, fine-tune, and deploy open-source AI models at high speed

Pricing model

Paid

Rating

★ People love it

Time to value

Quick Setup (< 1 hour)

You can use Fireworks AI to access and run open-source large language models and image models through a fast inference API. It lets you fine-tune models on your own data, deploy them to production, and run reinforcement learning training workloads. You can choose serverless pay-per-token pricing, dedicated on-demand deployments, or reserved capacity. It supports OpenAI and Anthropic-compatible APIs, making it easy to switch from existing setups without rewriting code.

What Fireworks does

Call open-source LLMs via REST API with a single line of codeFine-tune a base model using your own dataset through the Fireworks training SDKDeploy a fine-tuned model to production immediately after trainingSet up dedicated GPU deployments for consistent low-latency inferenceMonitor token usage and costs per model across projectsRun reinforcement learning training jobs on Fireworks GPU infrastructureIntegrate AI model inference into existing apps using OpenAI-compatible API formatCompare multiple open-source models side by side using the model libraryServerless pay-per-token pricing with no upfront contractsFull-parameter model fine-tuning including reinforcement learning loopsEvery trained model checkpoint deploys to production in secondsMulti-LoRA support for deploying multiple fine-tuned adapters on one base modelOpenAI and Anthropic API-compatible endpoints for easy migrationOn-demand dedicated deployments with multi-region supportAccess to latest open-source models like DeepSeek, Llama, Mistral, Gemma, and QwenIndustry-leading throughput and latency with custom inference engine