18 Hrs
Source:
شبكة راية الإعلامية
Source: شبكة راية الإعلامية
18 Hrsشبكة راية الإعلامية
Source: شبكة راية الإعلامية
18 HrsReady to play
Ready to play
A test involving 10 AI models and 273 coding tasks found that the top performer, “GPT-6 Astra,” met an average of 69.9% of the acceptance criteria, but fully completed tasks in only 26.4% of cases. Changes to interconnected code repositories were more difficult than multimodal tasks: the leading model completed 17.9% of the former, compared with 41% of the latter. The researchers concluded that models still struggle to complete complex projects without human intervention, and that performance is affected by the runtime environment as well as by the model’s capabilities.
Notice: This Is an AI-Generated Summary
Comments (0)