According to initial tests, GPT-5.5 is on par with Claude Mythos.
ioda/Shutterstock.com
A recent analysis by the UK’s AI Security Institute (AISI; part of the UK’s Department of Science, Innovation and Technology) shows: Open AI’s GPT-5.5 achieves a similar level to Anthropic’s Claude Mythos in advanced cybersecurity tasks. This would mean that it would be a threat comparable to that of the Claude creators’ model.
GPT-5.5 and Claude Mythos in the AISI comparison
For their investigation, the experts carried out 95 tests in the “Capture the Flag” format. In areas such as reverse engineering, web security and cryptography, Claude Mythos (in the preview version) solved 68.6 percent of the most difficult tasks. GPT-5.5 even managed to overcome an average of 71.4 percent of the challenges. As the AISI experts emphasize, this difference is within the margin of error.
GPT-5.5 was also successful in certain advanced tasks that previous models could not handle. For example, when simulating an attack on a company network within a test environment, the OpenAI model achieved comparable results to Claude Mythos’ preview version.
