Kimi K3 crashes the Fable 5 vs GPT-5.6 party at 1/4 the price

Moonshot's Kimi K3 just turned the top-model race into a three-way fight. The 2.8-trillion-parameter open-weight model — the largest ever shipped — went live July 16, with full weights due by July 27.

On the Artificial Analysis Intelligence Index it scores 57, nipping at Claude Fable 5 (60) and GPT-5.6 Sol (59) — a three-point spread that's basically a rounding error at the frontier. And it took #1 on the Frontend Code Arena at 1,679, beating both. Fable still edges agentic tasks (5 of 8 benchmarks); GPT-5.6 Sol leads research-grade physics.

The real story is the price tag. K3 runs $3 in / $15 out per million tokens vs Fable 5's $10 / $50 — roughly a 3x discount for near-identical output. It ships with reasoning always on ("thinking mode") and a 1M-token context. The-decoder notes it also signals the end of dirt-cheap Chinese AI: K3 is priced like a serious contender, not a loss-leader.

Why it matters: open weights at frontier quality and a quarter of the cost reset the price everyone else can charge.