Fireworks AI Inference Platform

What it does

Fireworks takes in prompts, images, audio, documents and text for embedding or reranking, and returns completions, structured JSON, function calls or vector embeddings from open models. It also accepts training datasets for supervised, preference and reinforcement fine-tuning, and produces custom models that can be served on the same platform. The docs list 100+ supported models across text, vision, audio, image and embeddings.

Teams typically start on pay-per-token serverless endpoints to test models, then move to dedicated GPU deployments for production, and fine-tune models on their own data. The API is described as a drop-in replacement for OpenAI for inference and training. What sets it apart is the combined stack: inference, model routing (ensembling) and training in one place, with single-endpoint access across 30+ regions, data residency options, and batch inference for asynchronous jobs.

It suits developers and engineering teams building AI applications who want open models instead of closed ones, and organizations needing region-aware deployment. Access requires an account and, for serverless, a payment method to buy prepaid credits. It is an API and platform rather than a consumer app, so you need to write code or use the CLI. Ensembling details and some pricing sit in separate docs pages.

What it can do

• Serverless inference on open models with per-token billing
• Dedicated GPU deployments with autoscaling
• Supervised, preference and reinforcement fine-tuning
• Function calling and structured JSON outputs
• Embeddings, reranking and batch inference
• Vision model support for images and documents

Works with

• Nexus CLI for coding tools

Example workflows

Prototype then deploy a chat feature

Inputs: user prompts and a chosen open model. Steps: sign up, call a serverless endpoint through the OpenAI-style API, test quality, then move to a dedicated deployment. Output: production completions served from dedicated GPUs.

Fine-tune a model on company data

Inputs: a dataset of conversations. Steps: upload data, start a supervised fine-tuning job via the UI or API, then deploy the tuned model. Output: a custom model served on the platform.

Before you choose

• Serverless billing is prepaid credits deducted by usage; spend limits and auto reload are available.
• Rates vary by model and serving path, so check the pricing docs before budgeting.
• Region-restricted deployments carry a 1.5x price premium.

Listed under

• Coding
• Sandboxes and infrastructure
• Developers
• Enterprise
• Fine-tuning
• GPU deployments
• Inference
• Inference API

Get started

Add a payment method to buy credits

1. Create an account at app.fireworks.ai
2. Add a payment method and buy prepaid credits
3. Pick a model using the model selection guide and call it via the API

Serverless is pay-per-token with prepaid credits; see the pricing documentation for per-model rates. A free Muse Glimmer 30B SFT offer runs until 11/30 (time-limited).

Source: fireworks.ai/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Own your specialized intelligence | Fireworks ↗fireworks.ai
• Fireworks - Pricing ↗fireworks.ai
• Build with Fireworks AI - Fireworks AI Docs ↗docs.fireworks.ai

Company website

fireworks.ai: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.