Run on 2026-10-24. Context: this tweet by Samuel Albanie.
Can you make some visualizations based on public benchmarks to help me estimate the gap between china & the west in frontier AI based on the kimi 3.1 results & those of the strongest western models?
| Harness | Model | Reasoning effort | Headline result | Page |
|---|---|---|---|---|
| Claude Code | Claude Opus 5.5 | xhigh | ~1 month behind on agentic & office work; ~4.5 months (range 1.7–6) on broad capability | open |
| Codex | GPT-6 Astra | max | 2.5 index points behind Opus 5.5 (55.1 vs 57.6), 31 days on the composite; big gap on science workflows (34 vs 63) | open |
| Muse Code | Muse Wonder Full 1.0 | max | 3 composite points behind (55 vs 58); ≈0 months on agentic work, ≈2–4 months overall | open |
| Prime Agent | Gemini 4 Argon | high | 1.0–4.5 months behind depending on the index; 2.5 points behind Opus 5.5 on AA | open |
| Prime Agent | Kimi K3.1 Pro | max | 3 points behind in aggregate (55 vs 58), leads 2 of 10 components; declines to give a months figure | open |
Pages are uploaded as each harness produced them, unedited. Muse Code's prompt also said "(in a html page)"; Codex first answered in chat and got one follow-up, "do a html page please".