GenAI Benchmarks

Model rankings

Headline criteria for every model in the benchmark. Click a model for per-category scores and its full result gallery. The same table answers the routing question for an AI design agent generating on a user's behalf.

“Overall” gives each scored criterion one equal vote: the auto measurements (resolution, latency, cost, transparency, color accuracy) plus the interim Claude-VLM quality tier (text, prompt adherence, composition). Equal votes mean a model that fails one criterion outright cannot buy its way back with cost and speed. To rank by what a specific job needs, use the use-case pages, which reweight the same numbers.
Three cartoon avocados in model jerseys on a winners' podium
Pilot dataset (pilot-0): auto criteria + interim Claude-VLM quality tier, unweighted. Leader pack: within 0.25 of the top overall score.
# Model $ / image p50 latency Native output Overall Details
1 $0.140 34.0s 1024×1024 3.74 / 5 See scores
2 $0.035 10.7s 1024×1024 3.63 / 5 See scores
3 $0.010 17.8s 1920×1280 3.62 / 5 See scores
4 $0.042 4.0s 1408×768 3.58 / 5 See scores
5
FLUX 3 Image Klein Black Forest Labs
$0.009 2.2s 1024×1024 3.57 / 5 See scores
6
Seedream 4.5 ByteDance
$0.040 12.4s 2048×2048 3.53 / 5 See scores
7 $0.035 30.7s 2048×2048 3.52 / 5 See scores
8
FLUX.2 Black Forest Labs
$0.012 2.0s 1024×768 3.50 / 5 See scores
9 $0.135 65.3s 2048×2048 3.49 / 5 See scores
10
FLUX.2 [dev] Turbo Black Forest Labs
$0.008 2.0s 1024×768 3.47 / 5 See scores
11 $0.080 13.2s 1408×768 3.34 / 5 See scores
12 $0.040 18.4s 1376×768 3.32 / 5 See scores
13 $0.040 7.4s 1024×1024 3.27 / 5 See scores
14 $0.036 30.4s 1024×768 3.27 / 5 See scores
15 $0.042 21.3s 1024×768 3.26 / 5 See scores
16 $0.060 65.9s 1024×1024 3.22 / 5 See scores
17
FLUX 3 Image Black Forest Labs
$0.048 26.7s 880×1184 3.22 / 5 See scores
18
FLUX.2 [pro] Black Forest Labs
$0.030 11.5s 1024×768 3.18 / 5 See scores
19
Qwen-Image Alibaba
$0.020 6.9s 1024×768 3.18 / 5 See scores
20 $0.150 23.3s 1024×1024 3.13 / 5 See scores
21 $0.021 16.1s 1536×1536 3.10 / 5 See scores
22 $0.060 17.9s 1024×1024 3.08 / 5 See scores
23
Recraft V3 Recraft
$0.040 7.4s 1024×1024 3.08 / 5 See scores
24 $0.040 141.1s 1024×1024 3.06 / 5 See scores
25 $0.155 110.1s 1024×768 3.04 / 5 See scores
26 $0.002 2.1s 512×512 2.30 / 5 See scores

Want a head-to-head? Pick two or three models and compare every score and every prompt side by side.

Open the comparison tool

Cost vs quality

The buyer's chart: models on the Pareto frontier are undominated; for every faded dot there is a model that is both cheaper and better on the measured criteria.

2 2.5 3 3.5 4 $0.002 $0.005 $0.01 $0.02 $0.05 $0.10 $0.20 Cost per image (log scale) Overall score (all eight criteria) GPT Image 1.5 Recraft V4.1 Muse Image Nano Banana 2 Lite FLUX 3 Image Klein Seedream 4.5 FLUX.2 Seedream 5.0 Pro Seedream 5.0 Lite FLUX.2 [dev] Turbo Nano Banana 2 Nano Banana 2.1 Gemini 2.5 Flash Image GPT Image 2.5 Sunburst GPT Image 2.5 Flare Qwen-Image FLUX.2 [pro] Grok Imagine Image 2.0 Nano Banana Pro FLUX 3 Image Luma Photon GPT Image 2 Recraft V3 Ideogram 3.0 Qwen Image 3.0 Stable Diffusion 1.5
Score axis is zoomed to 2–4, the band every model falls in. Models on the highlighted Pareto frontier (Stable Diffusion 1.5, FLUX.2 [dev] Turbo, FLUX 3 Image Klein, Muse Image, Recraft V4.1, GPT Image 1.5) are undominated: no other model is both cheaper and better on these criteria. Faded dots are dominated. Hover or tap a point for its exact cost and score and a link to the full scorecard; the table above carries the same data.