Stay free if you only need basic features. Upgrade if you need advanced features. Most solo builders can start free.
Fireworks provides access to a wide catalog of popular open-source models including Llama 3.1 (8B, 70B, and 405B), Llama 3.3 70B, DeepSeek V3, Qwen 2.5 (7B, 32B, and 72B), Gemma 2 (9B and 27B), Mixtral 8x22B, Mistral variants, and multimodal models like Llama 3.2 Vision. The library includes over 50 serverless models spanning LLMs, vision models, and image generation models like SDXL, with new models added frequently and often on launch day.
Fireworks uses per-token pricing that varies by model size and capability. Smaller models like Llama 3.1 8B are available at lower per-token rates, while larger models like Llama 3.1 405B cost more per token. A free tier is available for experimentation. Serverless endpoints require no upfront cost or GPU provisioning fees. On-demand dedicated GPU deployments are available for production workloads requiring guaranteed capacity. Enterprise customers can negotiate volume discounts with committed spend agreements.
Yes. Fireworks is SOC2, HIPAA, and GDPR compliant, offers zero data retention policies, and supports bring-your-own-cloud deployments for complete data sovereignty. Enterprise customers include Notion, Sourcegraph, Cursor, and Quora. The platform provides dedicated support, SLAs, and globally distributed infrastructure for mission-critical workloads.
Yes. Fireworks offers fine-tuning with advanced techniques including reinforcement learning, quantization-aware tuning, and adaptive speculation. You can customize any supported open-source model for your specific use case and deploy the tuned model directly on the Fireworks inference cloud without managing separate training and serving infrastructure.
Start with the free plan — upgrade when you need more.
Get Started Free →Still not sure? Read our full verdict →
Last verified March 2026