Chubby♨️@kimmonismus · XOpenAI的GPT-5.5 Cyber在网络安全能力上迅速缩小与Claude Mythos的差距，耗时仅数周而非数年。在AISI的专家网络任务中，两者表现接近，GPT-5.5 Cyber通过率甚至略高，且每token成本显著更低。但Mythos在公开实践案例上仍占优势，如协助Mozilla进行大规模Firefox漏洞排查。2026年正成为OpenAI的强势回归之年，其模型性能更强、成本效益更高，且一系列决策时机精准，展现出强劲复苏态势。
The surprising part is not just that Claude Mythos is powerful. It is that OpenAI seems to have closed much of the cyber-capability gap with GPT-5.5 Cyber in weeks， not years.
On AISI's expert cyber tasks， GPT-5.5 Cyber was roughly on par with Mythos and even slightly ahead on pass rate， while being materially cheaper per token. But Mythos still has the stronger public real-world proof point： Mozilla's large-scale Firefox vulnerability work.
Be that as it may， 2026 increasingly looks like OpenAI's comeback year： stronger releases， more cost-effective models， and a series of decisions that seem to be landing at exactly the right moment.