AI Agent Builders Tools
PRODUCT OVERVIEW
Fireworks AI is a developer platform for building and running generative AI on open models. It combines managed training, including supervised and preference fine-tuning, with an inference engine that serves the latest open models or your own trained versions. Developers can start serverless with per-token billing and OpenAI- and Anthropic-compatible APIs, move to dedicated on-demand GPU deployments for higher throughput, or reserve guaranteed capacity. A model library provides single-line access to popular open-source models, and training options range from guided runs to configuration-led jobs and custom training logic. It suits engineering teams building code assistance, conversational AI, agentic systems, search, multimodal, and enterprise RAG workflows. Access is self-serve via signup, with enterprise and reserved-capacity options available through sales contact.
Sign up at fireworks.ai/signup to access the dashboard, model library, and API.
Start with serverless per-token inference, or configure on-demand or reserved deployments for dedicated capacity.
Select a model from the library and call it through the OpenAI- or Anthropic-compatible API.
Use managed supervised or preference fine-tuning, or the Serverless Training API, to customize open models.
Add a payment method to buy credits, enable Auto Reload, and set a monthly spend limit for serverless usage.
Code assistance
Conversational AI
Agentic systems
Search
Multimodal applications
Enterprise RAG
It is a developer platform for training, fine-tuning, and serving open models, combining managed training with an optimized inference engine.
Serverless inference uses pre-paid, usage-based billing per token, with usage deducted from your balance and billed on output, input, and cached tokens.
Yes. You can enable Auto Reload to top up automatically when your balance runs low, or set a monthly spend limit to cap usage.
Serverless inference is described as OpenAI and Anthropic compatible, so you can call models through those API conventions.
Options include guided runs, configuration-led training, and writing your own training logic, plus managed supervised and preference fine-tuning.
On-demand deployments are billed per GPU second with no extra charges for start-up times; H100 and H200 GPUs start at $8.00 per GPU hour.
The pricing page lists a limited-time free offer for Muse Glimmer 30B SFT via the UI and Serverless Training API until 11/30, with no credit card required; other usage is paid.
Contact the Fireworks team through the contact or reserved-capacity pages to discuss enterprise deployments with faster speeds, lower costs, and higher rate limits.
Serverless inference uses pre-paid, usage-based billing per token. Training is priced per 1M training tokens or per GPU hour. On-demand deployments are billed per GPU second. Prices shown are starting rates in USD.
Pay-per-token inference with zero setup and no cold starts, billed on input, output, and cached tokens.
Supervised and preference fine-tuning priced per 1M training tokens, starting at LoRA SFT for models up to 16B parameters.
LoRA training on a shared, always-on trainer pool with no provisioning and no idle cost; you pay only for prefill, sample, and train tokens.
Dedicated GPU deployments billed per GPU second with no extra charges for start-up times, starting with H100 and H200 GPUs.
COMMUNITY REVIEWS
No reviews yet.
Anthropic's developer console and API for building agents and applications with Claude models and managed agent infrastructure.
AWS platform for building generative AI applications and agents at production scale with model choice, customization, guardrails and cost optimization.
iFlytek Open Platform provides APIs and solutions for large models, speech recognition, speech synthesis, OCR, and machine translation.
