Fireworks AI
The fastest platform for open-source AI
Last reviewed Mar 14, 2026
Fireworks AI is an inference and training platform for open-source generative AI models. Founded by the creators of PyTorch, it serves text, vision, audio, image, and embedding models through a pay-per-token API, dedicated GPU deployments, and supervised or reinforcement training paths.
LLM API Pricing
Pay-per-token pricing. Prices shown per 1M tokens.
Prices last updated: October 25, 2026
| Model | Input/1M | Output/1M |
|---|---|---|
| $0.150 | $0.600 | |
| $0.150 | $0.500 | |
| $0.300 | $1.20 | |
| $1.40 | $4.40 |
Fireworks AI pricing by model
Input, output, and batch rates, plus alternatives, for one model at a time.
Pros & Cons
Advantages
- Lightning-fast inference with industry-leading response times
- Easy-to-use API with OpenAI and Anthropic compatibility
- Wide variety of optimized open-source models
- Discounted pricing for cached tokens and batch inference
- Enterprise reliability with 99.99% uptime SLA
- Up to 100 fine-tuned models deployable without extra costs
Limitations
- Limited capacity with some serverless model limits
- Primarily focused on language models over image/video generation
- BYOC only available for major enterprise customers
- Feature-rich interface can have steep learning curve
Key Features
100+ Open-Source Models
Text, vision, audio, image, and embedding models, including recent releases such as Kimi K3, GLM 5.2, DeepSeek V4, MiniMax M3, and Qwen3.7 Plus
Blazing Fast Inference
Industry-leading throughput and latency with fast inference engine
Fine-Tuning Suite
SFT, DPO, and reinforcement fine-tuning of models up to 1T+ parameters, from guided runs through writing your own trainer and RL loop
OpenAI- and Anthropic-Compatible API
Drop-in replacement for closed-model APIs - change the base URL to migrate, with the same SFT data format for fine-tuning
On-Demand GPUs
H100, H200, B200, B300, and GB300 deployments with per-second billing and autoscaling
Batch Processing
Async bulk inference jobs run at scale for less than standard serverless requests
Embeddings and Reranking
Embedding and reranking models for search and context retrieval, billed per 1M input tokens
Fireworks Nexus
Drop-in replacement for closed-model APIs that routes each task to the best open or closed model, targeting 50-75% lower AI coding spend
Complete Model Lifecycle
Build, tune, and scale AI models without managing infrastructure
Pricing Options
| Option | Details |
|---|---|
| Serverless Inference | Pay-per-token pricing across Standard, Priority, and Fast serverless tiers with postpaid billing and no cold starts |
| Cached tokens | Cached input tokens billed at a reduced rate on supported models |
| Batch processing | Async bulk inference billed at a discount to standard serverless requests |
| Embeddings | Embedding models billed per 1M input tokens, tiered by model parameter count |
| Fine Tuning | Supervised and preference fine tuning billed per 1M training tokens with LoRA or full-parameter options; reinforcement fine tuning billed per GPU hour |
| Serverless Training API | Shared always-on trainer pool for LoRA training with no provisioning or idle cost, billed per token prefilled, sampled, and trained |
| On-demand GPUs | Per-second billing for H100, H200, B200, B300, and GB300 GPU deployments with no start-up charges; region-restricted deployments in the US or Europe are priced at 1.25x the standard rate |
| Reserved capacity | Guaranteed capacity with higher rate limits and earliest access to new hardware, arranged through sales |
Availability & Support
Regions
18+ global regions across 8 cloud providers with multi-region deployments and BYOC support for enterprise
Support
Documentation, Discord community, status page, email support, and dedicated enterprise support with SLAs
Getting Started
- 1
Explore Model Library
Browse 100+ models at fireworks.ai/models
- 2
Test in Playground
Experiment with prompts interactively without coding
- 3
Generate API Key
Create an API key from user settings in your account
- 4
Make first API call
Use OpenAI-compatible endpoints or Fireworks SDK
- 5
Scale to production
Transition to on-demand GPU deployments for production workloads
Frequently Asked Questions
Which models does Fireworks AI serve?
Check the pricing table above for the models Fireworks AI currently serves and their per-token rates.
How do I get started with Fireworks AI?
Explore Model Library, Test in Playground, Generate API Key, Make first API call, Scale to production
What are Fireworks AI's main advantages?
Fireworks AI's main advantages include: Lightning-fast inference with industry-leading response times, Easy-to-use API with OpenAI and Anthropic compatibility, Wide variety of optimized open-source models, Discounted pricing for cached tokens and batch inference, Enterprise reliability with 99.99% uptime SLA, Up to 100 fine-tuned models deployable without extra costs.
What are Fireworks AI's limitations?
Fireworks AI's main limitations include: Limited capacity with some serverless model limits, Primarily focused on language models over image/video generation, BYOC only available for major enterprise customers, Feature-rich interface can have steep learning curve.
Compare Providers
Find the best prices for the same GPUs and models from other providers