Kimi K3: The Largest Open-Weight Model Yet, at 2.8 Trillion Parameters

On 16 July 2026, Chinese company Moonshot AI released Kimi K3 — the largest open-weight model to date, with roughly 2.8 trillion parameters. It succeeds the Kimi K2 family, and full weights are due July 27.

The headline numbers

• 2.8 trillion parameters in a Mixture-of-Experts (MoE) design, but each token activates only 16 of 896 experts — about 1.8% — keeping inference far lighter than the total suggests.
• A 1-million-token context window, native vision, and always-on reasoning.
• Aimed at long-horizon coding and agent workloads.

Benchmark performance

In blind developer testing, Kimi K3 ranked #1 in the Frontend Code evaluation at 1,679 points, ahead of Claude Fable 5. It jumped 17 places from K2.6 (#18 → #1) with a 76% pairwise win rate — versus 63% for Claude Fable 5 and 58% for GPT-5.6 Sol. It marks a milestone in open models closing the gap with top U.S. frontier systems.

Architectural innovations

Kimi K3 introduces KDA — a hybrid linear-attention mechanism that interleaves linear-attention layers with periodic full-attention layers in a 3:1 ratio — plus "Attention Residuals," a drop-in replacement for residual connections designed to scale more efficiently.

API pricing

Moonshot listed pricing at $0.30 per million cache-hit input tokens, $3 per million on cache misses, and $15 per million output tokens. Opening the weights lets enterprises self-deploy rather than depend on an API.

Why it matters

Kimi K3 reinforces a 2026 trend: the gap between open and frontier commercial models on everyday work is narrowing to single-digit percentages, while costs run several times lower. For engineering teams, it is one more strong option to weigh when building AI products.

References

• Tom's Hardware — Moonshot releases 2.8-trillion-parameter Kimi K3
• VentureBeat — China's Moonshot AI releases Kimi K3
• Marc Pope — Kimi K3: 2.8T Parameters, #1 at Code