W3BStation
Markets
BTC $96,420 +2.34% ETH $3,280 +1.82% SOL $185.40 -0.92% BNB $642.50 +0.45% XRP $2.18 +3.12% DOGE $0.082 -1.50% ADA $1.05 +0.80% AVAX $42.10 +1.15%
BTC $96,420 +2.34% ETH $3,280 +1.82% SOL $185.40 -0.92% BNB $642.50 +0.45% XRP $2.18 +3.12% DOGE $0.082 -1.50% ADA $1.05 +0.80% AVAX $42.10 +1.15%
06/24/2026

OpenAI's GPT-5.5-Cyber Outperforms Anthropic Mythos on CyberGym Benchmark

What happened: OpenAI's GPT-5.

OpenAI's GPT-5.5-Cyber Outperforms Anthropic Mythos on CyberGym Benchmark

What happened: OpenAI's GPT-5.5-Cyber achieved an 85.6% score on the UC Berkeley-developed CyberGym cybersecurity benchmark, overtaking Anthropic's Claude Mythos 5 (83.8%). Anthropic's top models remain offline due to a Trump administration export ban imposed June 12 over national security and jailbreak concerns. GPT-5.5-Cyber, launched June 22, is accessible only to verified security professionals and is being deployed in partnership with CrowdStrike, Cisco, and Cloudflare.

Why it matters: The regulatory suspension of Anthropic's models has given OpenAI a temporary edge in the cybersecurity AI race, though the performance gap is narrow. The episode highlights how policy, not just technology, is shaping the competitive landscape in advanced AI, with access controls and export restrictions now central to model deployment strategies.

Source: Decrypt, The Decoder