CLAIM-2026-08-18-05
Terminal-Bench 3.0 rose from 4.6 to 28.3.
Why it matters
Shows the jump is not confined to security work — general agentic terminal capability moved too, on the same unchanged base.
Sources
- z.ai/blog/glm-5.3
- MarkTechPost
- Unite.AI
Terminal-Bench 3.0 rose from 4.6 to 28.3.
Shows the jump is not confined to security work — general agentic terminal capability moved too, on the same unchanged base.