Start on pay-per-token, reserve throughput as you scale, or run Fireworks inside your own cloud. Explore Platform for systems and apps or Nexus for code and co-work.
Start on shared serverless capacity, guarantee it as traffic grows, or run Fireworks inside your cloud.
Pay per token · From $0.05 / 1M input tokens
Call any serverless model in seconds. No commitment. Usage draws down prepaid credits.
Best for: Launching fast and spiky traffic
Commitment: No
Reserve dollars per minute · Serverless rates, reserved by the minute
Guaranteed throughput on serverless and on-demand models, backed by an SLA.
Best for: Steady production traffic
Commitment: Yes
Contract · Talk to us
Private preview for Enterprise
Fireworks-managed inference on GPUs in your cloud account or data center.
Best for: Data residency and existing GPU capacity
Commitment: Yes
Every serverless request bills on input, cached input and output tokens. Switch modes per request, on the same endpoint.
The default for most traffic.
Prioritize your traffic during peak load times.
100+ tokens per second for interactive apps.
New: Available for specific models, in preview. Talk to us.
50% off input and output for jobs that don’t need an immediate answer.
30% off Standard on best-effort capacity, available for select models only. Throughput can drop to zero, so retry or fall back to Standard.
USD per 1M tokens.
| Model | Input | Cached input | Output | Priority (input / cached / output) |
|---|---|---|---|---|
| Ember-1 | $3.00 | $0.30 | $15.00 | $3.75 / $0.375 / $18.75 |
| Kimi K3 | $3.00 | $0.30 | $15.00 | $3.75 / $0.375 / $18.75 |
| Qwen 3.8 Max | $2.00 | $0.25 | $6.00 | $3.00 / $0.375 / $9.00 |
| GLM 5.3 | $1.40 | $0.26 | $4.40 | $1.75 / $0.325 / $5.50 |
| Muse Spark Open | $1.25 | $0.15 | $4.25 | $1.875 / $0.225 / $6.375 |
| NVIDIA Nemotron 3 Ultra (Preview) | $0.60 | $0.12 | $2.40 | $0.75 / $0.15 / $3.00 |
| DeepSeek V4.1 Flash | $0.30 | $0.006 | $1.20 | $0.375 / $0.0075 / $1.50 |
| DeepSeek V4.1 Flash 1017 | $0.30 | $0.006 | $1.20 | $0.375 / $0.0075 / $1.50 |
| MiniMax M3 | $0.30 | $0.06 | $1.20 | $0.45 / $0.09 / $1.80 |
| GLM 5.3 Flash | $0.15 | $0.03 | $0.50 | $0.1875 / $0.0375 / $0.625 |
| OpenAI GPT OSS 120B | $0.15 | $0.015 | $0.60 | $0.18 / $0.018 / $0.72 |
| NVIDIA Nemotron 3.5 Lightning 30B A3B | $0.05 | $0.01 | $0.20 | $0.0625 / $0.0125 / $0.25 |
USD per 1M tokens.
US-only Serverless is available on select models at a 50% premium (waived on Muse Spark Open); see pricing here. Models not listed here are priced by size.
| Model | Tier | Cached input | Uncached input | Output |
|---|---|---|---|---|
| GLM-5.3 | Standard | $0.26 | $1.40 | $4.40 |
| GLM-5.3 | Spot | $0.182 | $0.98 | $3.08 |
| Kimi K3 | Standard | $0.30 | $3.00 | $15.00 |
| Kimi K3 | Spot | $0.21 | $2.10 | $10.50 |
| DeepSeek V4.1 Flash | Standard | $0.006 | $0.30 | $1.20 |
| DeepSeek V4.1 Flash | Spot | $0.0042 | $0.21 | $0.84 |
| DeepSeek V4.1 Flash 1017 | Standard | $0.006 | $0.30 | $1.20 |
| DeepSeek V4.1 Flash 1017 | Spot | $0.0042 | $0.21 | $0.84 |
| Base model | Price |
|---|---|
| Up to 150M parameters | $0.008 |
| 150M to 350M parameters | $0.016 |
| Qwen3 8B | $0.10 |
Token price is one input. Cache hits, the mode you choose and the model each turn goes to decide the rest.
Session affinity keeps each agent on the server that already holds its context. Long runs reach cache hit rates up to 99.6%.
Async jobs run at half the Standard rate.
Take 30% off for traffic that can retry or fall back
Send routine work to open models and save premium models for the turns that earn it.
Train at the rates below, then serve a fine-tuned model on Serverless or a Dedicated Deployment. See our inference rates or talk to us about a custom deployment.
In the UI and the Serverless Training API. No credit card required.
From $0.50 / 1M training tokens
No-code, just bring a dataset
Best for: ML engineers
By model · see table below
Customize your own training loop
Best for: ML researchers
From $8.00 / GPU-hour
You run the loop, at full scale
Best for: Advanced research teams
Priced per engagement.
Priced per 1M training tokens. Estimate training tokens as dataset tokens times epochs. Reasoning traces add tokens, because multi-turn conversations unroll into separate turns.
Full-parameter training is available on select models. For most full-parameter work, use the Dedicated Training API.
| Base model | LoRA SFT | LoRA DPO | Full Param SFT | Full Param DPO |
|---|---|---|---|---|
| Up to 16B parameters | $0.50 | $1.00 | $1.00 | $2.00 |
| 16.1B to 80B | $3.00 | $6.00 | $6.00 | $12.00 |
| 80B to 300B | $6.00 | $12.00 | $12.00 | $24.00 |
| Over 300B | $10.00 | $20.00 | $20.00 | $40.00 |
Billed per 1M tokens you prefill, sample and train. Repeated prefixes bill 80% off.
| Base model | Context | Prefill | Cached prefill | Sample | Train |
|---|---|---|---|---|---|
| Kimi K3 | 192K | $10.87 | $2.17 | $27.11 | $32.55 |
| Muse Spark Open | 256K | $6.19 | $1.239 | $15.48 | $18.58 |
| GLM 5.3 | 256K | $4.86 | $0.972 | $12.15 | $14.58 |
| GLM 5.3 Flash | 195K | $2.96 | $0.593 | $7.41 | $8.89 |
| Muse Glimmer 30B | 128K | $1.96 | $0.39 | $4.88 | $5.86 |
| Qwen 3.8 27B | 128K | $1.86 | $0.372 | $5.595 | $4.103 |
| DeepSeek V4 Flash 0731 | 256K | $1.74 | $0.35 | $4.33 | $5.20 |
| DeepSeek V4.1 Flash 1017 | 256K | $0.60 | $0.12 | $3.60 | $4.98 |
| NVIDIA Nemotron 3.5 Super VL 120B | 256K | $0.57 | $0.114 | $1.44 | $2.32 |
| Qwen 3.6 35B A3B | 256K | $0.45 | $0.09 | $1.335 | $2.14 |
Billed per GPU-hour at the GPU rates below.
Save with a monthly plan, or talk to us about Enterprise.
Prices in USD. Taxes may apply.
On Demand bills per GPU-second and the Dedicated Training API per GPU-hour, from the same rates. No charge for On Demand start-up time.
Idle deployments scale to zero after one hour by default. Region-restricted deployments are priced at 1.5x. Reserve GPUs, typically for one year, for guaranteed capacity, higher quotas, lower hourly rates and early access to new regions and hardware.
| GPU | Per minute | Per hour |
|---|---|---|
| H100 80 GB | $0.134 | $8.00 |
| H200 141 GB | $0.134 | $8.00 |
| B200 180 GB | $0.217 | $13.00 |
| B300 288 GB | $0.250 | $15.00 |
| GB300 288 GB | $0.334 | $20.00 |
Fireworks is a first-party provider in Microsoft Foundry, so usage counts toward your Microsoft Azure Consumption Commitment.
Discounts are available for bulk and prepaid purchases. Contracted customers can move to post-paid invoicing.
Set monthly spend limits, and tag requests by team, project and environment for chargeback.
Zero data retention by default. SOC 2 Type II, ISO 27001, 27701 and 42001, and HIPAA with a BAA.
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.
Yes. Set a monthly spend limit on your account. Auto Reload is separate: it buys more credits when your balance runs low, so production keeps running, and it never raises your limit.
Caching is automatic, with no setup and no write fee. Cached input bills at 50% off by default, with deeper discounts on select models.
We customize inference to your needs. Usage above your reservation can bill at standard Serverless rates, never a penalty rate. Overage isn’t covered by the SLA and follows your Serverless rate limits.
When utilization is high, when rate limits get in your way, or when you want to serve a model that isn’t on Serverless, including one you trained. On Demand bills per GPU-second, so a busy deployment usually costs less per token.
No. You pay for the deployment by the GPU-second, plus the cost of the training job.
By contract. Sovereign is available in preview to Enterprise customers. Talk to us to scope your environment.
Get in touch to discuss your goals, workloads, and find the best deployment option for your business.