Fireworks AI
Inference, routing and fine-tuning platform for open AI models
Own Fireworks AI? Claim it, free to correct the facts and get the verified badge.
Pricing
Paid
API
Yes
Company
Fireworks AI, Inc.
Platforms
What is Fireworks AI?
Fireworks AI is a platform for running, ensembling and training open models such as Kimi, DeepSeek, GLM, Qwen and gpt-oss. Developers call models through one endpoint available in 30+ regions. They can start on pay-per-token serverless capacity and later reserve throughput or deploy dedicated GPUs.
The inference engine is built for throughput, which the company puts at up to 4x that of providers wrapping vLLM, and for heavy prompt caching. Every serverless request can run in a Standard, Priority, Fast, Batch or Spot mode. Ensembling routes each turn to a model based on accuracy and cost. Training covers no-code managed SFT/DPO/ORPO, a Python Serverless Training API that supports RL, and dedicated GPU training.
Fireworks fits engineering teams and enterprises that want to move off closed frontier APIs, cut cost per task, or build domain-specific models. It offers data residency options, zero data retention by default, and SOC 2, ISO and HIPAA compliance.
Key features
- Serverless inference for open LLM, vision and embedding models
- Batch mode at 50% off, Spot mode at 30% off
- Automatic prompt caching with up to 99.6% hit rates
- Model ensembling that routes each turn by cost and accuracy
- Managed SFT, DPO and ORPO fine-tuning with LoRA
- Python Serverless Training API with RL support
- On-demand H100, H200, B200, B300 and GB300 GPUs
- Monthly spend limits and request tagging for chargeback
Use cases
Who it's for
Replacing closed model APIs
Run open models like DeepSeek or Kimi for workloads where they meet the quality bar, paying per token.
Fine-tuning domain models
Upload a dataset, launch LoRA SFT or DPO, then serve the tuned model on Serverless or a dedicated deployment.
Cost-sensitive background jobs
Send async work to Batch at half price or to Spot capacity at 30% off.
Regulated deployments
Use regional data residency or Sovereign deployments inside your own cloud with zero data retention.
Fireworks AI pricing
From the vendor's site, checked Nov 8, 2026; we re-check prices monthly. Prices change — confirm on their pricing page.
Serverless
pay per token
From $0.05 / 1M input tokens
- Prepaid credits, no commitment
- Standard, Priority, Fast, Batch and Spot modes
- Cached input at half price or less
Reserved Throughput
reserved by the minute
Custom
- Throughput SLA and 99.9% uptime terms
- Overage billed at standard Serverless rates
Fireworks Sovereign
contract
Contract
- Private preview for Enterprise
- Managed inference on GPUs in your cloud or data center
Managed Training
per token
From $0.50 / 1M training tokens
- SFT, DPO or ORPO from UI, CLI or API
- LoRA or full-parameter on select models
Dedicated Training API
per GPU-hour
From $8.00 / GPU-hour
- Full-parameter training
- Largest MoE models, longer context
Get an email when Fireworks AI changes its prices
One mail per change, a stop link in each. Nothing else.
How Fireworks AI compares
- Fireworks AI has no free plan, while 1 of the 3 alternatives below has one.
| Tool | Starts at | Free plan | Free trial | API | Open source | |
|---|---|---|---|---|---|---|
| FFireworks AI | — | — | — | — | ||
T Together AI | — | — | — | — | Fireworks AI vs Together AI | |
G Groq | — | — | — | — | — | Fireworks AI vs Groq |
O OpenRouter | — | — | — | Fireworks AI vs OpenRouter |
The closest alternatives in our directory by category. “Starts at” is the cheapest paid plan per month, as each vendor lists it.
Pros and cons
Strengths
- Serverless pricing starts at $0.05 per 1M input tokens with no commitment
- Cached input billed at half price or less, with no write fee
- SOC 2 Type II, ISO 27001/27701/42001 and HIPAA with BAA
- Usage counts toward Microsoft Azure Consumption Commitment
Limitations
- Spot capacity can drop to zero throughput
- US-only Serverless carries a 50% premium on select models
- Region-restricted GPU deployments are priced at 1.5x
- Sovereign deployment is in private preview
Fireworks AI FAQ
How does Fireworks billing work?
You buy credits and usage draws them down: per token for Serverless, per GPU-second for On Demand, and per token or GPU-hour for training.
Can I cap my spending?
Yes. You can set a monthly spend limit. Auto Reload buys more credits when your balance runs low but never raises the limit.
Is there anything free?
Muse Glimmer 30B SFT training is free through November 30 in the UI and Serverless Training API, with no credit card required.
Does Fireworks keep my data?
Zero data retention is the default, and region-specific data residency is available.
How much do GPUs cost?
H100 and H200 cost $8.00/hour, B200 $13.00, B300 $15.00 and GB300 $20.00, billed per GPU-second for On Demand.
Used Fireworks AI? Rate it
Sign in to rate this tool and share how you use it.
Fireworks AI
Paid