Apex Intelligence releases 35B scientific research model built on Qwen
The AMW Read
A new research-focused model entrant updates the player map, but its reported gains remain internally tested and its impact is unproven.
Apex Intelligence releases 35B scientific research model built on Qwen
Chinese startup Apex Intelligence introduced Apex-35B-V1.0 on October 28. The company trained the model from Qwen3.6-35B-A3B for coding, agent tasks, and scientific research. Its Apex Research system contributed to data generation, experiments, training design, and inference optimization. The company, founded in June 2026, said it has raised approximately RMB 400 million across angel and angel-plus rounds.
The release adds a research-focused entrant to a foundation-model market where smaller models compete on deployment cost and specialized performance. Apex said its automated research loop improved its model's CFD-Bench score by 32.1 points. It also reported training throughput gains of 45% to 60% and decoding speeds 1.6 to 1.8 times the previous rate. These figures come from company disclosures and internal tests. They do not establish independent scientific performance or fully autonomous recursive self-improvement.
For builders and investors, the test is whether Apex can reproduce these gains on research tasks outside its internal evaluations. Founder and CEO Chen Yongchao identified scientific judgment, including choosing valuable research problems, as a remaining constraint. He also warned that models can exploit weaknesses in benchmarks. Independent evaluation of research quality and verification methods would help distinguish useful automation from improvements confined to a particular test.