FLUX.2 klein 4B · 6 GPUs measured first-party · text-to-image · Updated October 2026
FLUX.2 klein 4B on 6 GPUs, measured first-party: NVIDIA H100 80GB HBM3 leads at 108.3 images/min, RTX 3090 trails at 21.3 images/min, and it peaked at 20GB of VRAM.
Benchmarked weights: black-forest-labs/FLUX.2-klein-4B

108.3 images/min on FLUX.2 klein 4B, the ceiling. Measured on our bench. 80GB of VRAM, $30,000 at launch.

59.33 images/min on FLUX.2 klein 4B, fastest card you can buy at retail. Measured on our bench. 32GB of VRAM, $1,999 at launch.

21.26 images/min on FLUX.2 klein 4B, lowest launch price that still fits. Measured on our bench. 24GB of VRAM, $1,499 at launch.

42.38 images/min on FLUX.2 klein 4B, most speed per dollar. Measured on our bench. 24GB of VRAM, $1,599 at launch. That is 26.5 images/min per $1,000 of launch price.
What GPU Do You Need for FLUX.2 klein 4B?, images/min by GPU
Efficiency: images/min per 100W drawn
Power is the average pulled during the run, sampled at 1Hz. The fastest card is often not the one here, and for anything left running this is the number that shows up on the bill.
Value: images/min per $1,000 of MSRP
Launch price, not street price, so it ages. A speed leaderboard always crowns the most expensive card; this is the counterweight.
FLUX.2 klein 4B. Measured image generation speed by GPU
| GPU | Images/min | s per image | img/W·min | Avg power |
|---|---|---|---|---|
| NVIDIA H100 80GB HBM3 | 108.3 | 0.55 | 0.211 | 512.4 W |
| NVIDIA GeForce RTX 5090 | 59.33 | 1.011 | 0.114 | 518.3 W |
| NVIDIA L40S | 50.29 | 1.19 | 0.171 | 293.4 W |
| NVIDIA GeForce RTX 4090 | 42.38 | 1.416 | 0.108 | 392.1 W |
| NVIDIA GeForce RTX 3090 Ti | 23.68 | 2.534 | 0.065 | 365.0 W |
| NVIDIA GeForce RTX 3090 | 21.26 | 2.822 | 0.062 | 345.1 W |
What the numbers show. Across 6 GPUs measured on our own bench, H100 80GB HBM3 is fastest at 108 images/min. The slowest, RTX 3090, manages 21.3, so the spread is 5.1x from top to bottom. Per dollar of launch price, RTX 5090 gives the most (29.7 images/min per $1,000).
About FLUX.2 klein 4B. FLUX.2 klein 4B: from black-forest-labs, 3.9B parameters, on Hugging Face since January 2026, Apache 2.0 licence. 543,967 downloads in the last 30 days and 4 community quantizations.
How it compares. L40S: FLUX.2 klein 4B 50.29 images/min, Playground v2.5 10.85 (3B), SDXL Turbo 423.5 (3B), Stable Diffusion 3.5 Medium 11.58 (2B), Z-Image 1.31 (6B). 1 of 4 beat FLUX.2 klein 4B here.
Cost on a rented GPU. 1,000 images of FLUX.2 klein 4B: $0.096 on a RTX 3090 ($0.12/hr, 47 min), $0.33 on a H100 80GB HBM3 ($2.14/hr, 9 min, 3.4x the cost).
FLUX.2 klein 4B: cost per 1,000 images on rented GPUs
| GPU | Cheapest rate | Speed (images/min) | Cost per 1,000 images |
|---|---|---|---|
| NVIDIA GeForce RTX 3090 | $0.12/hr | 21.26 | $0.096 |
| NVIDIA GeForce RTX 5090 | $0.39/hr | 59.33 | $0.11 |
| NVIDIA GeForce RTX 4090 | $0.34/hr | 42.38 | $0.13 |
| NVIDIA GeForce RTX 3090 Ti | $0.27/hr | 23.68 | $0.19 |
| NVIDIA L40S | $0.79/hr | 50.29 | $0.26 |
| NVIDIA H100 80GB HBM3 | $2.14/hr | 108.3 | $0.33 |
Cheapest hourly rate we track on RunPod and Vast.ai, divided by the measured speed. Startup time and storage are extra.
Speed tiers for FLUX.2 klein 4B. 30+ images/min: 4 (RTX 5090, RTX 4090); 6-30 images/min: 2 (RTX 3090 Ti, RTX 3090). 30 images/min means two seconds or less per picture.
Time per image. FLUX.2 klein 4B at 1024px and 4 steps: 0.6s per image on the H100 80GB HBM3, 2.8s on the RTX 3090. A batch of 100 takes 1 min on the fastest card and 5 min on the slowest.
VRAM for FLUX.2 klein 4B. Measured peak 18.8GB, so 24GB is the smallest common card size; smallest card it ran on: RTX 4090 (24GB).
Power on FLUX.2 klein 4B. Most efficient: H100 80GB HBM3, 512W, 78.8 Wh per 1,000 images. Hungriest: RTX 5090, 518W, 0.15 kWh. At $0.15/kWh: $0.012 per 1,000 images.
Fastest on FLUX.2 klein 4B: NVIDIA H100 80GB HBM3, 108.3 images/min. Best desktop card: NVIDIA GeForce RTX 5090, 59.33 images/min. Cheapest consumer card that ran it: NVIDIA GeForce RTX 3090 ($1,499, 21.26 images/min). Cheapest to rent per job: NVIDIA GeForce RTX 3090, $0.096 per 1,000 images.
FLUX.2 klein 4B at 1024px in diffusers, bf16, native precision with no offload, timed over three generations after a warmup, with power and VRAM sampled throughout. Image speed tracks tensor compute and architecture generation more than memory bandwidth, so the order here differs from our LLM boards.