GPT-5 for Vision: Results from 80+ Real-World Tests

On August 7, 2025, OpenAI introduced GPT-5, a model in their GPT series that combines advanced reasoning abilities with multimodal support, allowing it to process both textual and visual inputs. GPT-5 demonstrated strong performance in reasoning tasks, ranking high on Vision Checkup, a tool for evaluating vision models, but showed mixed results in areas such as object counting and defect detection. The model was successful in some document understanding and OCR tasks, yet struggled with precise object measurement and detection in complex scenarios, achieving a lower mAP50:95 score on the RF100-VL benchmark compared to the current state-of-the-art Gemini 2.5 Pro. The introduction of reasoning capabilities marks a significant advancement in the field of multimodal models, even though issues like the stochastic nature of responses and initial testing flaws were noted. Despite these challenges, GPT-5's ability to integrate reasoning into visual tasks suggests a promising future for models that analyze images with greater insight.