Start free →

Model APIs

Fireworks AI

Watched pages: pricing ↗

Current pricing

How Fireworks AI charges

Fireworks AI offers usage-based pricing for serverless inference and training, with on-demand deployments billed per GPU second or hour.

Serverless Inference (Embeddings up to 150M)$0.008per 1M input tokens
  • Pay per token
  • Zero setup
  • No cold starts
  • Usage-based billing
Serverless Inference (Embeddings 150M - 350M)$0.016per 1M input tokens
  • Pay per token
  • Zero setup
  • No cold starts
  • Usage-based billing
Serverless Inference (Qwen3 8B)$0.1per 1M input tokens
  • Pay per token
  • Zero setup
  • No cold starts
  • Usage-based billing
Managed Training (Models up to 16B parameters)$0.50per 1M training tokens (LoRA SFT)
  • Supervised and preference fine tuning
  • Minimal setup
On Demand Deployments (H100 80 GB GPU)$8.00per hour
  • Pay per GPU second
  • No extra charges for start-up times
  • Faster speeds
  • Higher rate limits
On Demand Deployments (B300 288 GB GPU)$15.00per hour
  • Pay per GPU second
  • No extra charges for start-up times
  • Faster speeds
  • Higher rate limits

Captured Oct 21, 2026 from the public pricing page.

Change log

1 reviewed change

P0pricing

Fireworks AI updates pricing model from postpaid to prepaid and introduces new model pricing

Fireworks AI has changed its billing model from postpaid to prepaid, requiring users to add a payment method and buy credits. The company also introduced specific pricing for new models like Muse Spark Open, DeepSeek V4.1 Flash, and GLM 5.3, and a limited-time free offer for Muse Glimmer 30B SFT.

Before · Sep 3, 2026Pay per token, with high rate limits and postpaid billing. Get started with $1 in free credits.
AfterPay per token with pre-paid, usage-based billing. Simply add a payment method to buy credits, and usage is then deducted from your balance.

Why it matters. This shift to a prepaid model could impact user acquisition and retention, as it requires an upfront commitment. The introduction of new model pricing and a free tier for a specific model indicates an expansion of their offerings and a strategy to attract users to new services.

Live source ↗Archived before ↗

Compare

Compare with

Fireworks AI vs DeepSeek APIFireworks AI vs GroqFireworks AI vs Mistral AIFireworks AI vs OpenAI APIFireworks AI vs Together AI