Data◆Article
Leading open weight models now trail closed frontier models on cybersecurity capability by only 4-7 months, down from the 6-10 month gap measured through most of 2025.
The UK AI Security Institute's first public delta analysis found that open models GLM-5.2 and DeepSeek V4-Pro are catching up to proprietary frontier models on narrow cyber capability evals faster than in prior periods. ✦ AI generated
UK AI Security Institute (AISI) · Import AI · 2026-07-20 · original ↗
Recent open models GLM-5.2 and DeepSeek V4-Pro perform similarly to frontier closed models released 4 to 7 months before them – a narrower gap than the 6 to 10 months we measured through most of 2025.
Read full article ↗excerpt · fair-use quotation
- ·Open models now trail closed frontier models by 4-7 months
- ·Down from 6-10 month gap seen through most of 2025
- ·GLM-5.2, DeepSeek V4-Pro match older closed frontier models
- ·First public delta analysis from UK AI Security Institute
- ·Narrower lag on narrow cyber capability evals
- ·Gap measured in months behind closed frontier models
- ·UK AI Security Institute (AISI)
- ·GLM-5.2 and DeepSeek V4-Pro cited as fastest catch-up cases
- ·Evals focus on narrow cyber capability, not general capability
Around this claim