Run open models, connect your coding tools, and train models for your own tasks.
Get started · Documentation · Model library · Community · Get help
| Your goal | Start here |
|---|---|
| Build an application | Python SDK: call models, stream responses, and work with the Fireworks API. |
| Connect your coding tools | Nexus CLI: use Fireworks in Claude Code, Codex, Cursor, and other coding tools. |
| Train a model for your task | Training cookbook: runnable recipes for fine-tuning, reinforcement learning, and distillation. |
| Measure inference performance | Benchmark suite: measure latency, throughput, and streaming performance. |
Training from your coding agent? Add training skills to help choose a method, plan a run, and debug jobs. Explore the Training API docs for more detail.
More tools: use firectl to manage deployments, models, datasets, and training jobs from your terminal. See the Nexus docs for coding-tool setup.
Follow a guided workshop from setup to a working result. Each workshop includes its own requirements and instructions.
- Train a research agent: use reinforcement learning for multi-hop research, then run the model as an agent.
- Train a model router: train a decision model to route between models.
- Turn a skill into model weights: use supervised fine-tuning and a Python scoring function.
Read MiniMax kernels: open-source GPU kernels for MiniMax models, including MiniMax M3 sparse attention on NVIDIA Blackwell. The design and performance write-up explains how they work.
- Questions, ideas, or something to share: join the Fireworks Community to discuss what you are building and learn from other developers.
- API and setup reference: visit the documentation.
- Repository bugs or feature requests: open an issue in the relevant repository. For bugs, include the steps to reproduce them.
- Account, billing, or private troubleshooting: contact Fireworks support.
- Want to contribute: follow the contribution guidance in the repository you want to improve.