Apex 35B

Apex-35B-V1.0 is the first model from Apex Intelligence (Apexin). It is based on Qwen3.6-35B-A3B and was further developed through mid-training and post-training using proprietary datasets and research trajectories. The vendor says AI took part in its development, including data construction, experiment design, code execution, training-result analysis and training optimization.

The model targets standard tasks such as coding, agent workflows, question answering and deep search. It is also aimed at end-to-end computational research workflows in AI research, computational science and robotics simulation. The vendor reports that it outperforms Qwen3.6-35B-A3B and Recursive Flint on six research benchmarks, and reports the highest BFCL V4 score among the models it compared. These are vendor-reported results.

Demonstrations on the website cover robotic task execution (RoboRSI, scored 44.5), real-to-sim scene reconstruction from a single photo (Real2Sim, scored 21.55), and CFD optimization (79% win rate against the base model on CFD-Bench). A YouTube creator reports the full model is about 72GB, with the smallest quantization about 9GB. The official page says further details on capabilities, benchmarks and availability will be announced later, so license and download locations are not confirmed in the provided material.