Microsoft Says MDASH Beats Claude Mythos and GPT-5.6 Sol in Cybersecurity Test

Microsoft has release its first dedicated cybersecurity version called MAI-Cyber-1-Flash and connects it to MDASH, a vulnerability detection system that claims to beat Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol while costing 50% less than Microsoft’s best MDASH configuration, according to the company.

The combined setup scored 95.95% on CyberGym, according to Microsoft. CyberGym is a benchmark that asks AI agents to reproduce 1,507 known vulnerabilities in 188 open-source projects, then finds them and the number of successful releases in controlled environments.

This put MDASH ahead of GPT-5.5 Cyber​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​in front of GPT-5.5 Cyber ​​​​​​​​at 85.6%, Mythos 5 at 83.8%, GPT-5.6 Sol at 83.6%, and Gemini 3.5 Flash Cyber ​​​​​​at 83.2%. The results were self-reported by Microsoft and did not appear in CyberGym’s public leaderboard at the time of publication, although the brand uses a public test and success metric.

MAI-Cyber-1-Flash does not work by itself. Microsoft says it works up to 90%, while MDASH uses 10% more difficult up to GPT-5.4. This is important because tokens – the artificial intelligence’s tokens – cost money every time a brand reads a code or responds.

This is the first time that Microsoft’s brand designed for efficiency can outshine the art that is created by the most advanced technology. “When combined with MDASH, (MAI-Cyber-1-Flash) delivers world-class performance at 50 percent of the cost of leading models,” Microsoft CEO Satya Nadella wrote.

The model (in this case MAI-Cyber-1-Flash) is an AI that thinks in code. A chain (MDASH in this case) is a circular machine: assistants, tools, checks, and workflows that decide where to look, challenge what is suspected, remove duplicates, and confirm that an error can cause it.

MDASH employs more than 100 specialized employees who are tasked with analyzing, debating whether the findings are real, and creating a proof of concept—a working demonstration that the error exists.

Microsoft said that early detection and delays can be eliminated because AI makes the discovery of the virus cheaper. The company says that many years of safety gives it an advantage, adding, “No one can make this record.”

Ever since the release of Claude Mythos, cyber security experts have been trying to beat or match their capabilities. Researchers have also created a Mythos threat scan with a sample of people for less than $30 per scan. Dawid Moczadło, one of the researchers involved, said that “moat is moving from the path of acceptance to certainty.”

The missing piece is becoming a system that validates the findings without burying the false alarm generators.

Decrypt also said that GPT-5.5 Cyber ​​had recently taken CyberGym ahead with 85.6% score, slightly beating Mythos. Microsoft’s score is 10 points higher than GPT-5.5 Cyber ​​​​and 7.5 points higher than MDASH’s results, but it measures the whole system and not MAI-Cyber-1-Flash itself.

Microsoft is putting MDASH in the secret image through Microsoft Security Exposure Management in the Defender portal. Customers can browse Git repositories, view uncommitted commits, and use the Defender CLI to generate code they want to view.

This preview reduces the database to about 256MB and allows for one scan per tenant. Project Perception is expected to expand the multi-agent simulation process beyond code analysis and threat analysis as well as operational management reform.