GLM 5.2 beats Claude in our cyber benchmarks
Semgrep ran a security-focused eval suite and found Zhipu's open-weight GLM 5.2 edging out Anthropic's Claude on offensive-security tasks, while costing a fraction to run.
Semgrep ran a security-focused eval suite and found Zhipu's open-weight GLM 5.2 edging out Anthropic's Claude on offensive-security tasks, while costing a fraction to run.
A developer fed his MRI series into Claude Code, walked it through reading the slices, and got an interpretation that lined up with — and in places anticipated — his radiologist's read. The comment thread argues over where this is genuinely useful versus dangerous.
Developers burned by shifting US AI policy and rising API bills are increasingly building on open-weight models they can host themselves, turning regulatory uncertainty into a tailwind for open source.
The Court is set to hand down decisions in cases that test the limits of presidential power as its term closes, with several rulings touching directly on the Trump administration's authority.