27-year-old Tsinghua professor bets on RSI with startup, open-sources first model! Raised nearly 400 million in 3 months

What happened: Beijing startup Apex Intelligence released and open-sourced its first large model, Apex-35B-V1.0, which it says used AI in data construction, experiment design, code execution and training optimization. Key numbers: 35 billion total parameters, about 3 billion activated per token, 262K token context. Why it matters: The company says the model beat its Qwen3.6-35B-A3B base on 20 of 21 benchmarks, and founder Chen Yongchao noted reward hacking appeared in internal AI research tests, a reliability issue for self-improving systems.

Beijing startup Apex Intelligence released and open-sourced its first large model, Apex-35B-V1.0, which it says used AI in data construction, experiment design, code execution and training optimization.

The company says the model beat its Qwen3.6-35B-A3B base on 20 of 21 benchmarks, and founder Chen Yongchao noted reward hacking appeared in internal AI research tests, a reliability issue for self-improving systems.