We’ve released Apex Flash-1 with Yeta Labs, an open-weight model built on GLM-5.3-Flash and post-trained for cybersecurity. As models take on more security investigations, teams have to account for the cost of finding and verifying a vulnerability. We want that work to be affordable enough to run regularly.
We built training environments around vulnerability cases from our security research, using running applications where the model has to investigate a flaw and demonstrate a working exploit. Apex Flash-1 is designed to work under a larger model that assigns it a specific task. It then reads code, uses tools, tests hypotheses, and checks its findings.
The model is built on the same research strategy that brought us to the number one spot on HackerOne’s US business leaderboard for 2026.
We’ll continue building training cases from our research and testing future versions against new cases, with the aim of improving how well the model performs security tasks while lowering its running costs.
