Breaking

Moonshot AI Releases 2.8-Trillion-Parameter Kimi K3, Topping a Key Coding Benchmark as China Chases the US

July 20, 2026 at 19:18 EDT

Foundation Models
Open Source
Software Dev & Coding

Beijing-based startup Moonshot AI has released Kimi K3, a 2.8-trillion-parameter model that trails the top American systems on overall capability but claims the No. 1 spot on a widely watched coding leaderboard, sharpening the debate over how quickly China can close the gap with the United States in artificial intelligence.

July 16, 2026 · Moonshot AI

Kimi K3 Grabs the Coding Crown — but Not the Overall Lead

China's Moonshot AI unveils a 2.8-trillion-parameter model that tops a closely watched coding leaderboard, beating Claude and GPT on code and agentic tasks — while still trailing the top US systems on aggregate capability.

2.8T

total parameters — one of China's largest models to date

1.8%

of the network active per token (16 of 896 experts)

token context window, with native vision support

on the Frontend Code Arena coding leaderboard

Frontend Code Arena — leaderboard score

A sharp jump over the prior generation lifts K3 past Claude into first place.

1,679

Kimi K3

🥇 No. 1

Claude

Fable 5

prior leader

Kimi

K2.6

previous gen

API pricing — per million tokens

input tokens

$15

output tokens

$0.30

cached input

Fable-class coding at a fraction of the cost — the price-to-performance ratio enterprises see as a genuine competitive threat.

What impressed reviewers

Top-ranked coding and agentic behavior
Single-prompt games, visualizations, decks
Kimi Work praised for document & codebase analysis
Aggressive price-to-performance ratio

Where it still falls short

Trails top two models on aggregate benchmarks
Merely adequate on some domain-specific tasks
Immature setup documentation
Loose safety guardrails, bypassable in a few turns

The next test

Full model weights arrive July 27, with a portion published as open-weight. The real question: whether K3's coding lead holds up outside curated benchmarks — and how far China's efficiency-driven, export-control-constrained strategy can carry it against better-resourced American rivals.

Continue reading

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