OpenAI's GPT-6 Astra vision model achieves state-of-the-art object detection performance, scoring 82.1% mAP@50 on Roboflow Vision Evals and outperforming competitors like Qwen3.8 Max and GPT-5.6 Sol. The model excels at computer vision tasks including object detection, visual reasoning, box prompting, and segmentation, combining fine-grained detail detection with semantic understanding for effective auto-annotation and classification across diverse visual scenarios.
A startup needed reliable drone detection that could distinguish drones from birds and planes while tracking their location. The engineer built a real-time system that outperforms published state-of-the-art models on all object sizes by using a modern detection backbone, adding a high-resolution head for small distant objects, and fixing corrupted training data that caused false positives.
A year-long project trained traffic-sign detectors using only synthetic rendered images, achieving 75–84% of the performance of models trained on real photographs. Fine-tuning the synthetic-pretrained model on real data improved detection by 0.15 mAP50, with the largest gains on underrepresented sign classes, though modeling snow proved ineffective due to lack of real snowy conditions in test data.