OpenAI unveils GPT‑5.5‑Cyber: it tops Claude Mythos 5 in cybersecurity

OpenAI has just unveiled the full release of its AI, GPT‑5.5‑Cyber. According to one assessment, the model now edges ahead of Anthropic’s Claude Mythos 5.

For several weeks, the press has repeatedly highlighted Claude Mythos, Anthropic’s powerful AI which, beyond its advanced level of intelligence, is particularly strong at identifying security vulnerabilities. However, as of the time of writing, Claude Mythos 5-still on hold due to a directive from the United States government-would no longer be the leading AI for cybersecurity work. This week, OpenAI presented the complete version of GPT‑5.5‑Cyber, a GPT‑5.5 variant tuned specifically for cyber use. And based on evaluation results shared by ChatGPT’s creator, this new AI would now sit ahead of Claude Mythos 5 in that specific area.

Benchmark results for GPT‑5.5‑Cyber on CyberGym

GPT‑5.5‑Cyber recorded a score of 85,6 % on the CyberGym benchmark, compared with 83,8 % for Mythos 5. This evaluation measures how capable AI models are at tasks linked to vulnerability analysis in software.

As expected, GPT‑5.5‑Cyber is not being offered to the general public. Like Anthropic, OpenAI is limiting its cyber-focused AI to defensive use cases, made available only to a restricted number of organisations.

Codex Security update and expanded Daybreak access

Alongside announcing the full version of GPT‑5.5‑Cyber, OpenAI also revealed an update to Codex Security, the cyber-specialist plug-in within its agent platform. The company also said it is widening access to its Daybreak programme. This programme enables organisations to use Codex Security, as well as GPT‑5.5 with Trusted Access for Cyber-another cyber-oriented AI, but one that does not match the same level as GPT‑5.5‑Cyber.

The timing is perfect

In any case, OpenAI is making these cybersecurity-related announcements at a moment when its rival, Anthropic, still cannot restore access to Claude Mythos 5 and Claude Fable 5. As a reminder, the US government cited national security reasons when asking Anthropic to block access to these two models for foreign nationals. To comply with the directive, Anthropic chose to suspend both AIs for everyone.

Discussions between Anthropic and the US government are already under way. And recently, after meeting Anthropic’s CEO at the G7 summit, Donald Trump said that he no longer views the AI lab as a national security threat.