FLUX.2 [dev] tops open-weights text-to-image on Artificial Analysis
- Source
- x.com
- Date
FLUX.2 just landed in @ArtificialAnlys and @arena. • FLUX.2 [dev] is the new leading open-weights model for text-to-image • FLUX.2 [pro] & [flex] are #2 overall • Frontier-level quality without frontier-level prices FLUX.2 [dev] has passed 1M downloads across the BFL and ComfyUI variants on Hugging Face. FLUX.2 [klein] is coming soon - built for consumer GPUs.
We worked with @nvidia_ai_pc & @ComfyUI to offer a lighter version of FLUX.2 [dev] with an FP8 quantization version. Read more here https://t.co/NsTda9r4hP
• Try FLUX.2 [pro] and [flex] on our playground https://t.co/BnGavpx4pi • FLUX.2 [dev] weights https://t.co/VgOIL7nHIE
Context
FLUX.2 dev's reported lead concerned the open-weights category. In its December 5, 2025 announcement, Black Forest Labs called dev the leading downloadable text-to-image model, while its accompanying leaderboard screenshot placed it ninth overall.
Those statements describe different comparison groups. Artificial Analysis uses blind human preferences to assess image quality. A model can lead the open-weights subset while other models rank above it overall. The announcement therefore supports a narrower conclusion about quality among downloadable alternatives at that date.
- quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.
Checking sign-in…
Loading comments…