Usman Salis
⬤ FLUX.2 Klein 9B from Black Forest Labs just grabbed the crown for best open-weights image editing model, according to fresh benchmark data from Artificial Analysis. The model scored an ELO rating of 1,158 based on thousands of real user preferences, putting it ahead of Alibaba's Qwen Image Edit 2511 (1,151) and even its bigger sibling FLUX.2 [dev] Turbo (1,149).
⬤ What makes this interesting is that the 9B version is actually beating Black Forest Labs' own 32B FLUX.2 Dev model in editing tasks, despite having way fewer parameters. The FLUX.2 Klein family is showing up strong across the board too - multiple variants from the lineup are sitting in the top ten spots on the leaderboard.
⬤ FLUX.2 Klein comes in four flavors across two sizes: 9B and 4B parameters, each with main and base versions. The main models use a four-step distillation process for quicker image generation, while the base versions work better if you want to fine-tune them yourself. All variants run under Apache 2.0 licensing, so you can use them commercially without restrictions. For straight text-to-image generation, the 9B variant ranks fourth among open models.
⬤ Price-wise, FLUX.2 Klein is pretty competitive. On Black Forest Labs' API, the 9B model costs $17 per 1,000 edited images (with 1MP input), while Qwen Image Edit 2511 runs you $60 for the same volume. Between the performance, open licensing, and lower costs, open-weights image models are becoming serious options for creative work and production pipelines.
Usman Salis
