On July 28, Microsoft AI unveiled its new self-developed cybersecurity model, MAI-Cyber-1-Flash. This model operates on the MDASH platform—a multi-agent framework capable of detecting and fixing vulnerabilities in large-scale codebases. According to MDASH, the combination of the MAI-Cyber-1-Flash and GPT-5.4 models scored 12 points higher than Mythos in the CyberGym benchmark test. Microsoft AI stated that the configuration of this model offered in MDASH is priced 50% lower than its previous optimal solution (GPT-5.4 + 5.4 mini + 5.3 codex).
All Comments