Benchmarks and production analysis of AI models — measured, not guessed.
In a blind test by 13 production professionals, Seedream 5 Pro led the frozen Qwen Image 2.0 outputs but trailed GPT Image 2 on believability. Agreement was low; prompt adherence did not separate Seedream from GPT.
20 min read
In a blind test of 48 commercial scenes, GPT Image 2.5 Sunburst clearly beats Qwen Image 2. Against GPT Image 2 it takes 52.7% of believability votes, which is not a clear win.
Qwen Image 2.0 Pro wins 47.0% of blind believability judgments against its predecessor (24.8% losses) and 19.7% against GPT Image 2 (50.6% losses).
A 48-scene blind evaluation of GPT Image 2, Qwen Image 2, FLUX.2 Klein 9B, and real photographs by 13 production professionals.
264 image pairs, 13 evaluators, 792 blind judgments. SeedVR2 took detail recovery 86% to 2% and naturalness 82% to 2%. The one scale where Magnific closes the gap is also the one our evaluators could not agree on.
In the research, 13 evaluators compared SeedVR2 and Topaz Wonder 3 on 264 image pairs and tied 54-60% of them. SeedVR2 leads on edge sharpness, Topaz Wonder 3 wins faces, and the free model is the defensible default.
SeedVR2 vs FLUX.2 [klein] 9B: blind evaluation of 264 image pairs by 13 professional evaluators across 4 quality dimensions. SeedVR2 wins naturalness at 91%, sharpness at 83%. Exception: AI-generated textures at close range.
Get new benchmarks and datasets by email when they are published.
jamie@example.com Subscribe