OpenAI's GPT Image 1.5 tops benchmarks but struggles with creative
OpenAI releases GPT Image 1.5 with faster generation, improved text rendering, and stronger benchmark scores, though user reception has been mixed relative to competing image models.
OpenAI released its new image model GPT Image 1.5, featuring precise image editing, better instruction following, improved text and markdown rendering, and faster generation up to 4×. Despite topping multiple leaderboards like LMArena (1277), Design Arena (1344), and AA Arena (1272), user feedback from Twitter, Reddit, and Discord communities is largely negative compared to Nano Banana Pro by Gemini. Xiaomi introduced the MiMo-V2-Flash, a 309B MoE model optimized for inference efficiency with 256K context window, achieving state-of-the-art scores on SWE-Bench.
The model uses Hybrid Sliding Window Attention and multi-token prediction, offering significant speedups and efficiency improvements. The timing of OpenAI's launch amid competition from Gemini and Nano Banana Pro affects user sentiment, highlighting challenges in benchmarking relevance.