Breaking

Moonshot AI Releases Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Challenging GPT-5.6 Sol

July 18, 2026 at 22:45 EDT

Foundation Models
Open Source
Software Dev & Coding

Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight model that the Chinese startup positions as a frontier-class rival to OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5, particularly in long-horizon coding and visual tasks. The model became available via API immediately upon its release around July 16, 2026, with full weights slated for public download on July 27.

July 16, 2026 · Moonshot AI

Kimi K3: China's Open-Weight Model Reaches Frontier Territory

A 2.8-trillion-parameter open-weight model that competes head-to-head with GPT-5.6 Sol and Claude Fable 5 — not by undercutting on price, but on quality. It marks the end of the era of super-cheap Chinese AI.

2.8T

parameters — first Chinese open model in the ~3T class

token context window, natively multimodal

16/896

experts active — high-sparsity Stable LatentMoE

Intelligence Index — competitive, not dominant

Independent benchmark score (higher is better)

Kimi K3

GPT-5.6 Sol

Claude Fable 5

Per-token price — K3 undercuts by 40–50%

USD per million tokens · headline API rates

Input

Sol

Output

$15

$30

Sol

Where K3 wins

Top-tier on WebDev & frontend arenas — beats Sol and Fable 5 in some cases
Strong one-shot UI, 3D modeling, large-repo work
Open weights (July 27) — self-host and fine-tune
40–50% lower per-token pricing

Where it lags

Lower token efficiency — burns more tokens per job
Slower responses; Sol often finishes faster
Real-world cost can end up equal or higher
Tighter usage limits reported

The real test: can an openly distributed model close the last gap to closed frontier systems?

The trade-off hinges on whether K3's lower pricing and openness outweigh its speed and efficiency disadvantages against GPT-5.6 Sol — a dynamic that intensifies both the U.S.–China race and the open-versus-closed debate.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.