Research
Moonshot AI and Tsinghua release TCSExam, a Lean-checked benchmark that sets AI agents 100 theoretical computer science research tasks
Researchers from Tsinghua University, Moonshot AI and the Shanghai Qizhi Institute published TCSExam on GitHub on November 3, 2026, with an arXiv paper two days later. GPT-6 Astra leads the 100-task Full suite at 71.9 an…