GPT-5.4 Pro has achieved a remarkable score of 90% on the EyeBench-V2 benchmark, a significant improvement over the next best model, which scored 71%, and far surpassing the scores of other leading models that range between 14-37%. This benchmark tests AI models on complex vision tasks that have historically posed challenges in the field, and the leap from 71% to 90% in just one generation highlights the rapid advancements in AI vision capabilities. Such progress reflects ongoing efforts toward developing superhuman vision in AI systems.
GPT-5.4 Pro: GPT-5.4 Pro is OpenAI’s premium AI model variant engineered for superior performance on complex tasks, integrating coding, reasoning, and computer use capabilities.4824 It is accessible via API for developers and in ChatGPT for advanced applications.43 In this news, it demonstrates a breakthrough in vision processing by leading on the EyeBench-V2 benchmark, surpassing competitors and nearing human performance levels.440
Model Leap: GPT-5.4 Pro advances vision understanding dramatically in one generation over prior leading scores.0
AI Progress: The result underscores accelerating improvements toward superhuman AI vision capabilities.20
Vision Benchmark: EyeBench-V2 tests AI models on challenging vision tasks considered a longstanding hurdle in the field.40
Source: TFTC21
