Breaking

Moonshot AI Unveils Kimi K3, a 2.8-Trillion-Parameter Open-Weight Model Narrowing the US-China AI Gap

July 18, 2026 at 08:04 EDT

Foundation Models
Open Source
Infra & Chips

China's Moonshot AI released Kimi K3 on July 16, 2026, a 2.8-trillion-parameter open-weight model it bills as the world's largest openly available AI system, intensifying questions on Wall Street about whether the technology boom's leaders can sustain their lead. The launch, which pairs frontier-level performance with freely downloadable weights, has become the latest signal that Chinese developers are closing the distance to top US systems.

July 16, 2026 · Moonshot AI

Kimi K3: the world's largest open-weight AI model

Beijing's Moonshot AI unveils a 2.8-trillion-parameter, natively multimodal system priced aggressively and free to download — narrowing the distance to top US models and reviving Wall Street's doubts about the AI trade.

2.8T

parameters — billed as the largest openly available model

~1M

token context window, flat-rate across the full window

16/896

experts activated at inference (Mixture of Experts)

1st

Frontend Code Arena — 1,679 pts, edging Claude Fable 5

Scaling efficiency vs the previous generation

K3's Stable LatentMoE architecture claims ~2.5× better overall scaling efficiency than K2.

1×

Kimi K2

2.5×

Kimi K3

Frontier-tier on public benchmarks

1,679

Frontend Code Arena — #1, ahead of Claude Fable 5

~57

Artificial Analysis Intelligence Index — near Opus 4.8 & GPT-5.5

Top

Vals Index — lands near the top tier

Pricing — per million tokens, flat across the full window

input · cache miss

$0.30

input · cache hit

$15

output

Strengths cited

Among the most capable open-weight systems
Strong front-end, web, UI/UX and 3D generation
Long-running agentic workflows; beats Fable on Next.js

Caveats noted

Still trails strongest closed systems in some areas
Heavy output-token use can push real costs above list price
Full weights not yet shipped — local testing limited

A capable, open-weight, aggressively-priced Chinese model challenges the assumption that frontier AI is a durable US moat.

Arriving after the DeepSeek shock, K3 has revived investor worries about chip stocks and the staying power of the AI trade. Full weights are due by July 27 — the coming weeks should reveal how far the gap has truly narrowed.

Continue reading

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