BeijingA 27-year-old professor’s startup open-sourced Apex-35B, built partly by AI.

Apex-35B scored 60.9 on PaperBench, above 753-billion-parameter GLM-5.3’s 42.5, the company reports.

Its automated research system generated training data, designed experiments and tuned training, lifting throughput 45% to 60%, the company says.

All benchmark scores come from the company’s own release.

Weights are public under Apache 2.0, so outside labs can now test the claims.

How each outlet framed it
Zhidx leans favorable
reports the AI-driven research loop: the model was built with AI doing data construction and training analysis, and 8 of 34 AI-written papers scored above 3.0 at ARR

Sources: Zhidx