Moonshot AI unveiled Kimi K3 at Shanghai's World AI Conference, a 2.8-trillion-parameter model with a 1.05-million-token context window that the company says is its most capable to date. On Artificial Analysis's GDPval-AA v2 benchmark -- which measures real-world task performance across 44 occupations and nine major industries -- Kimi K3 scored 1,687, placing third overall behind Claude Fable 5 Max (1,815) and GPT-5.6 Sol Max (1,747.8), and ahead of Claude Opus 4.8 (1,600). On AA-Briefcase, a private agentic benchmark testing long-horizon knowledge work, K3 climbed to second place at 1,527, beating GPT-5.6 Sol Max's 1,495. It took the No. 1 spot outright on Arena.AI's Frontend Code Arena.
Moonshot attributes roughly a 2.5x improvement in scaling efficiency over its prior Kimi K2 model to two architectural changes: Kimi Delta Attention, a hybrid linear-attention scheme, and Attention Residuals, which change how information moves between transformer layers -- efficiency gains rather than pure parameter-count scaling, echoing the efficiency-first approach that made DeepSeek's earlier releases notable.
“AI infrastructure spending has outrun what's actually needed to stay competitive.”
Pricing is aggressive relative to frontier U.S. models: $0.30 per million cache-hit input tokens, $3 per million on cache misses, and $15 per million output tokens, available now via API with open weights promised by July 27. That combination of near-frontier benchmark performance and a fraction of the price is precisely what triggered the broader chip-stock selloff this week, on fears it signals U.S. AI infrastructure spending has outrun what's actually needed to stay competitive.
For developers and enterprises, Kimi K3's open-weight release later this month will matter more than the benchmark scores alone -- it gives any company the ability to self-host a near-frontier model rather than depend on API access from a small number of U.S. labs. What to watch next: independent benchmark verification once the open weights ship, and whether Western enterprises adopt K3 given ongoing data-sovereignty and export-control considerations around Chinese-origin models.