Kimi K3 ships — the largest open-weight model to date lands #1 on one board, #3 overall

Why it matters (spine — moat). The scarce asset as intelligence goes free. Treat frontier-class capability as available open-weight and self-hostable; build every garden feature on a model-swap harness, not a single vendor's API.

What would prove it wrong (falsifier). The 2026-07-27 weight release slips indefinitely, or an independent eval shows K3 was benchmark-gamed and is not near-frontier in the wild.

Live-status check + triangulation. Moonshot AI's Kimi K3 — a 2.8T-parameter multimodal MoE with a 1M-token context — went live 2026-07-16 with full open weights promised by ~2026-07-27, priced $3/$15 per M tokens and shipped in MXFP4 (not ternary). Name the board: it is #1 on the Frontend Code Arena (1,679, ahead of Fable 5's 1,631) and tops several sub-domains, but trails Claude Fable 5 and GPT-5.6 Sol on overall capability — #3 on the Pareto frontier, not first. Triangulated across CNBC, Bloomberg, VentureBeat, Tom's Hardware, HuggingFace and Simon Willison (2026-07-16/17). Source: Model card · Moonshot AI (via HF) (2026-07-16); see source link. Every post-cutoff figure was tool-fetched against a live source during the Kimi K3 sweep (2026-07-20); see sweeps/2026-07-20-VERIFIED_DIGEST.md for the full provenance and any corrections of record.