Benchmarks

1 post tagged “Benchmarks”.

A red 'NO HACK' stamp over a download icon, showing the cyber ability was deliberately left out of the open release.

ai7 min read

Kimi K3.1 Pro's open weights are released with cyber ability removed

Moonshot released Kimi K3.1 Pro's open weights with cybersecurity capability deliberately reduced. The checkpoint scores 74.7 on ExploitBench versus 100 for GPT-6 Astra, while coding stays strong at 81.0 on xdev-bench-hard. Students reach the model via API, not the 1.56 TB download.

5 November 2026Arnab Biswas

Browse every post