Fireworks AI pricing: GPU rental rates
Visit Fireworks AI →Updated July 2026 · day 14 of the ledger · verified 2026-07-18
The cheapest GPU on Fireworks AI right now is the H100 (unspecified) at $7.00/hr (serverless, verified 2026-07-18). Fireworks AI is an AI inference platform. Most of its pricing is per-token serverless model pricing, which is out of scope for a GPU-hour tracker; what we track is its on-demand deployments table, where you rent dedicated GPUs for your own model deployment, billed per GPU-second with no extra charge for start-up times. Current hardware covers H100, H200, B200 and B300. There is no VM or SSH access, so we label these rates serverless in our tables.
Live Fireworks AI prices (2026-07-18)
| GPU | Tier | $/hr | Verified |
|---|---|---|---|
| H100 (unspecified) | serverless | $7.00 | 2026-07-18 |
| H200 (SXM) | serverless | $7.00 | 2026-07-18 |
| B200 | serverless | $10.00 | 2026-07-18 |
| B300 | serverless | $12.00 | 2026-07-18 |
Cheapest listed per-GPU-hour rates verified at the source. Before storage, egress and networking.
Pricing model
On-demand deployments are billed per GPU-second at a published hourly rate per GPU type, and Fireworks states there are no extra charges for start-up times. This buys a dedicated GPU serving your chosen model on the Fireworks platform rather than a virtual machine. The same page prices serverless inference per token and fine-tuning per training token; those are model-level prices, not GPU rentals, so our tracker deliberately excludes them.
Hardware lineup
The on-demand deployment table currently lists H100 80GB, H200 141GB, B200 180GB and B300 288GB GPUs. Fireworks also documents reinforcement fine-tuning jobs priced per GPU-hour at the same rates as on-demand deployments.
Who should pick Fireworks AI
Pick Fireworks on-demand deployments if you want a dedicated GPU serving your model with per-second billing and platform-managed infrastructure, especially for production inference where serverless per-token pricing stops making sense at volume. It is not a fit for general-purpose GPU compute, training your own stack with custom images, or anything needing root access; for raw GPU rentals, compare the VM providers in our tables. A formal public uptime SLA for on-demand deployments was not found on the pages reviewed. Not publicly documented.
FAQ
No. On-demand deployments provide dedicated GPUs running your model on the Fireworks platform, billed per GPU-second. There is no VM or SSH access, so we label the rates serverless.
The current table lists H100 80GB, H200 141GB, B200 180GB and B300 288GB.
Most of the page prices serverless inference per token and fine-tuning per training token. Those are model prices, not per-GPU-hour rental prices, so only the on-demand GPU deployment table qualifies for our index.
Profile verified 2026-07-18. Sources: fireworks.ai · fireworks.ai