Doubleword measures the open-weights to closed-source LLM gap across 18 benchmarks

Category: AI
Status: discussion
Sources: Doubleword blog, discussion
Summary: Jamie Dborin (Doubleword) argued on 2026-06-22 that the apparent convergence of open-weights and closed-source models depends heavily on the benchmark chosen. Extrapolating a single Artificial Analysis Intelligence Index trend implies the gap closes around 2026-12-03, but averaging across 18 benchmarks the gap stays nearly flat at just under 5 months over the period. Coding showed the largest narrowing, from about 15 months behind to one or two months. The analysis uses public Artificial Analysis data; the visualizations are not reproducible from the text alone.
Why it matters: It cautions against reading a single benchmark as proof open weights have caught closed models, while confirming coding is the fastest-closing axis.