AI coding agents fully complete only 26.4 percent of real-world tasks
GPT-6 Astra tops Alibaba's new Co-Design Bench at 26.4 percent full task completion. Harness choice swung one model's score by 19 points, making paired testing essential.
Topic
Topic
GPT-6 Astra tops Alibaba's new Co-Design Bench at 26.4 percent full task completion. Harness choice swung one model's score by 19 points, making paired testing essential.