fal pricing in 2026: Model APIs, Serverless GPU and Dedicated Compute explained
Nutzungsbasierte Plattform für generative Medien mit über 1,000 produktionsreifen Modellen und flexibler GPU-Skalierung, ideal für Entwickler, die KI-Funktionen ohne feste Verpflichtungen einbauen.
Starting price
$1.89 (Dedicated Compute (H100))
Free option
No
Plans
4
Prices checked
Sep 29, 2026
Dedicated Compute (B300)$4.49
Dedicated Compute (H100)$1.89
fal plans and what they include
fal has no subscription tiers. You choose a way to consume it and pay for what you use, with no minimum spend or commitment. Model APIs are billed per output, from $0.0002 to $1.00 depending on the model and what it produces (video duration, image resolution), across 1,000+ models for video, image, audio and 3D, with no monthly or per-call caps.
For your own models there are GPUs. Serverless H100 is $1.89 per hour, billed per second of execution and $0 when idle, and scales from 0 to thousands of GPUs. Dedicated Compute is billed for the full hour: H100 (80GB) at $1.89 per hour and B300 (288GB VRAM) at $4.49 per hour, both with full SSH access. Volume discounts and annual commitments go through sales. Prices checked on 2026-09-29.
fal plans compared
Plans and prices
Plan
Price
Billing
Who it is for
Key limits
API
Model APIs (Pay-Per-Use) API
Nutzungsbasiert
Per API call
Building generative media features with minimal fixed overhead
No per-month or per-API-call caps; 1,000+ models available; Pricing varies by model and output type (e.g., video duration, image resolution)
Included
Serverless GPU (H100) API
$1.89 per hour (billed per second of execution)
Hourly + per-second execution
Deploying custom or fine-tuned models with automatic scaling
Auto-scales from 0 to 1,000s of GPUs; Billed per-second when running; $0 when idle; Minimum execution ~1 second per request
Included
Dedicated Compute (B300) API
$4.49 per hour (288GB VRAM)
Hourly
Training, fine-tuning, and sustained compute workloads with guaranteed performance
Full SSH access (no autoscaling); Billed for full hour regardless of usage; Dedicated instance (no sharing)
Included
Dedicated Compute (H100) API
$1.89 per hour (80GB VRAM)
Hourly
Single-GPU development and fine-tuning at lower cost than B300
Full SSH access; Billed hourly minimum; Single-GPU instance
Included
Which fal plan to pick
Which plan to pick
If you
Pick
Why
you want to add image, video or audio generation to an app without upfront spend
Model APIs (pay-per-use)
No minimum or commitment; price per output varies from $0.0002 to $1.00 by model.
you deploy a custom or fine-tuned model and traffic is spiky
Serverless GPU (H100), $1.89/hr
Billed per second of execution, $0 when idle, auto-scales from 0, with private model deployment.
you train or fine-tune on a single GPU
Dedicated Compute (H100), $1.89/hr
80GB VRAM, full SSH access and persistent storage; billed hourly with a minimum.
you run sustained, heavy training workloads
Dedicated Compute (B300), $4.49/hr
288GB VRAM, InfiniBand for multi-GPU setups and multi-region availability. Billed for the full hour regardless of usage.
What is free, what costs extra
Free access and trial
The vendor does not describe a free tier on the pages we read.
Extra costs and taxes
Gebühren externer Zahlungsdienstleister; bei großen Ausgaben können Kosten für ausgehenden Datenverkehr anfallen
Costs that are easy to miss
There is no flat subscription and no cost cap unless you enforce one, so bills scale with usage and can be unpredictable at high volume.
Some models use GPU-based pricing, so inference for GPU-heavy models can cost more than alternative SaaS pricing.
Third-party payment processor fees apply, and egress bandwidth charges may apply for large outputs.
An account, API key and payment method on file are required before you can use it.
fal Preise und Nutzungslimits
Preise am Sep 29, 2026 auf der offiziellen US-Preisseite geprüft. Offizielle Preise ↗ Preise und Limits ändern sich häufig.
Preisseite des Anbieters vor 12 Tagen automatisch neu geprüft: keine Änderungen festgestellt.
Die offizielle fal-Preisseite, wie wir sie gesehen haben. fal.ai, erfasst Oct 1, 2026
Kostenpflichtige Tarife: was sich beim Upgrade ändert
Für: Single-GPU development and fine-tuning at lower cost than B300
Limits
Full SSH access
Billed hourly minimum
1 mehr
Single-GPU instance
Enthält
80GB VRAM H100 GPU
Full SSH control
2 mehr
Persistent storage for code and data
Development and fine-tuning optimized
API inklusive
Nutzungsobergrenzen, Guthaben und Zusatzkosten
Gebühren externer Zahlungsdienstleister; bei großen Ausgaben können Kosten für ausgehenden Datenverkehr anfallen
How much fal costs per year
At the listed prices, a single user pays between $23 and $54 a year on the fixed-price plans. Annual billing discounts, taxes and usage-based extras are listed per plan above where the vendor publishes them.
fal FAQ
Does fal have a free plan?
No free tier or trial is stated in our verified data. Model APIs have no minimum spend or commitment, but you need an account and a payment method.
How much does fal cost per image or video?
It varies by model and output, from $0.0002 to $1.00 per output. Video duration and image resolution affect the price.
What does an H100 cost on fal?
$1.89 per hour, both serverless (billed per second of execution) and dedicated (billed hourly).
What is the difference between serverless and dedicated compute?
Serverless auto-scales and charges only while running. Dedicated gives you a full instance with SSH access, billed for the full hour regardless of usage.
Can I get a discount on fal compute?
Volume discounts and annual commitments are available through contact sales.
More about fal
Full fal guide: features, use cases, how-to and sources.
