OPEN LINK ↗
// 1h agoBENCHMARK RESULT
Epoch Cyber ECI Shows Open Models Losing Ground
Epoch AI’s Cyber ECI places Z.ai’s GLM-5.3 at the top of open-weight models with a score of 160, ahead of GPT-5.4 but behind newer closed models. That puts open cyber capabilities roughly six to seven months behind the closed frontier.
// ANALYSIS
Open models remain highly capable, but the cyber gap is becoming strategically important as closed models advance faster.
- –GLM-5.3’s score comes from only three CAISI benchmarks, so the ranking carries meaningful uncertainty.
- –Closed models retain a preparation window before comparable cyber capability becomes broadly available.
- –Developers prioritizing self-hosting and auditability may accept weaker frontier performance for greater control.
- –Epoch cautions that benchmark coverage is uneven and earlier scores are not directly comparable after the October refit.
// TAGS
cyber-ecibenchmarkevaluationsecurityopen-weightsllmresearch
DISCOVERED
1h ago
2026-10-17
PUBLISHED
1h ago
2026-10-17
RELEVANCE
9/ 10
AUTHOR
0x0SojalSec
