Rolling record / ARQ 78 Severe · ▲ +24 7-day 314 events in window / as of 14 Sept 2026
ARQAI Risk Quotient
capability 83 Danger

Cognition Releases SWE-2: A Kimi K3 Post-Trained Coding Model That Matches Fable 5.1 on FrontierCode at 64% Lower Cost

A cheaper coding model matching frontier performance lowers barriers to capable AI-assisted software development.

A cheaper coding model matching frontier performance lowers barriers to capable AI-assisted software development.

Auto-scored: A cost-efficient frontier-matching coding model is a real but incremental capability advance with potential for minor misuse.