Every factual gag in this duel comes from public reporting:
Moonshot's table for Kimi K3.1 Pro vs Claude Opus 5.5: Kimi leads BrowseComp 92.1 vs 90.8, SWE Marathon 51.7 vs 46.3 and GPQA-Diamond 94.2 vs 93.9; Opus leads ExploitBench 91.0 vs 74.7, Terminal-Bench 4.0 66.4 vs 65.7 and BenchCAD 96.2 vs 92.7 — the-decoder.com(2026-10-23)
Moonshot calls Kimi K3.1 Pro its most capable 3T-class open-weight model, with a 1M-token context window and a 2-3x faster Highspeed variant; full weights to be released by Nov 4, 2026 — kimi.ai(2026-10-23)
Launch-day agent test (one small task, 3 runs each): Kimi K3.1 Pro at max effort took 28 steps, 12.2K reasoning tokens and 4.9 min; Claude Opus 5.5 took 7 steps, 0.7K reasoning tokens and 69 s; both passed 96/96 hidden tests. List prices per 1M tokens: Kimi $2 in / $10 out, Opus 5.5 $4 / $20 — atomicagent.io(2026-10-23)
Kimi K3.1 Pro launched 4th on the Artificial Analysis Intelligence Index (v4) with 55, behind Claude Opus 5.5 at 58; Kimi K3 had scored 44 — artificialanalysis.ai(2026-10-23)
Moonshot published the Kimi K3.1 Pro weights (2.8T total, 104B active parameters) on Hugging Face — tokenpost.com(2026-11-05)
Moonshot-reported Terminal-Bench 4.0: Kimi K3.1 Pro 65.7% vs Claude Opus 5.5 66.4%; BenchCAD 92.7% vs 96.2% — cryptobriefing.com(2026-10-23)
Kimi K3.1 Pro has roughly 2.8 trillion total parameters with 104 billion active (mixture of experts) — startupfortune.com(2026-10-23)
'You're absolutely right!' became a running joke about Claude Code, including a GitHub bug report with hundreds of reactions — github.com(2025-07-12)