Skip to content
Tier B — Production
Runs in:EUMade in:Germany
Hugging Face (nscale)

FLUX.1 Schnell

Tier B — Production

Tokonomix Editorial Team·Reviewed by Mes Kalkan·
Section 01

Speed analysis

Latency measured across all benchmark runs. P50 (median) and P95 (95th percentile) give a realistic picture of response speed under normal and peak load.

P50 latency (median)P95 latency12 runs
31641250860470008-1908-21ms
Section 02

Pricing

What one image costs you on Tokonomix, per size and quality.

Price per image — FLUX.1 Schnell

from $0.0045 per image

Billed per image. The price depends on size and quality; your prompt is included.

SizeQualityPer image
Every image$0.0045
Section 03

Tokens per second

Throughput in tokens per second, derived from measured P50 latency. Higher is better; fluctuations track provider-side load.

Throughput (tokens / s)320 / avg 423
625272

Estimated from P50 latency × 200 output tokens — the absolute number depends on this assumption; the trend is what matters.

Section 04

Capabilities

modalities: image
Section 05

Availability

Availability

No measurements yet

We haven't recorded enough API calls to show availability stats for this model. Data appears once the model starts receiving live traffic.

Section 06

Tokonomix benchmark verdicts

No benchmark verdicts yet for this model.

Section 07

Full model profile

FLUX.1 [schnell]: the fast, open-weight tier built for volume, not the leaderboard

FLUX.1 [schnell] is the smallest and fastest member of Black Forest Labs' FLUX.1 family, released alongside FLUX.1 [dev] and FLUX.1 [pro] on August 1, 2024. "Schnell" is German for fast, and the name is literal: the model is distilled specifically to produce a usable image in one to four diffusion steps, where most contemporaries need twenty or more. It is also the only tier in the original FLUX.1 lineup shipped under a fully permissive open-source license, which is why it shows up wherever teams want image generation they can run themselves rather than call through someone else's API.

Where it shines

Speed is the entire premise. Fewer steps means lower latency and lower compute per image, which matters for any workflow generating images in bulk — thumbnails, product mockups, placeholder art, iteration loops where a designer is cycling through dozens of variants before picking one. The model is also genuinely open: weights are published on Hugging Face, and nothing about running it depends on a hosted API staying available.

Under the hood

FLUX.1 [schnell] is a 12-billion-parameter rectified flow transformer, trained with latent adversarial diffusion distillation to collapse the normal multi-step diffusion process down to a handful of steps. Black Forest Labs describes the result as competitive prompt following and output quality for a model this fast, though it is explicitly positioned below [dev] and [pro] on raw fidelity.

Prompting notes

Because the model commits to an image in very few steps, prompts benefit from being direct rather than layered with qualifiers the model has no steps left to act on. Overloaded prompts with many competing instructions tend to produce a rushed compromise rather than any one instruction done well. It is a text-to-image model — there is no built-in editing or masking endpoint in this tier, unlike Black Forest Labs' later Kontext line.

Where it falls short

Black Forest Labs is direct about the limits in the model card: it does not provide factual information, can amplify biases present in its training data, and does not always match what was asked for. Compositional accuracy on complex multi-subject scenes lags the [pro] tier, and fine detail — hands, dense text, intricate patterns — is the first thing to degrade when a prompt asks for too much at once. The model card also lists out-of-scope uses, such as exploitative, non-consensual or identity-harming content.

Rights and licensing

Black Forest Labs publishes the FLUX.1 [schnell] weights under the Apache 2.0 license, which permits personal, research and commercial use with attribution and without royalties. The model card separately lists uses Black Forest Labs does not want the model used for; read both before building on it. That combination — permissive commercial license plus open weights — is uncommon among image models of comparable quality; most competitive text-to-image models are either closed API-only products or open under non-commercial terms.

When to pick it

Reach for [schnell] when volume and latency matter more than top-end fidelity: bulk asset generation, rapid prototyping, any pipeline where a human reviews and regenerates rather than expects a first-shot final image. It is also the right starting point for teams that specifically want to self-host rather than depend on a vendor API.

Alternatives worth comparing

Within the same catalogue, gemini-2.5-flash-image and gpt-image-1-mini cover the same fast/light end of the market from hosted vendors, with editing built in where [schnell] has none. For teams that need higher fidelity and are willing to trade self-hosting for it, gpt-image-1 sits closer to the [pro]-tier end of the quality range. Our EU-hosted models listing is the relevant filter for teams whose constraint is data residency rather than raw speed.

Last technical review: 2026-09-14 — Tokonomix.ai

Last automated test
Aug 21, 2026 · 20:01 UTC · Speed benchmark
P50 latency
625 ms
P95 latency
666 ms
Errors
3 / 6 runs
Last reviewed by Tokonomix Team·September 14, 2026