Breaking

Moonshot AI Launches 2.8-Trillion-Parameter Kimi K3, Topping Anthropic's Claude Fable 5 on Frontend Coding Benchmark

July 19, 2026 at 17:45 EDT

Foundation Models
Open Source
Software Dev & Coding

Moonshot AI released Kimi K3 on July 16, 2026, a 2.8-trillion-parameter open-weight Mixture-of-Experts model that immediately edged past Anthropic's Claude Fable 5 to claim the top spot on the Frontend Code Arena, sharpening the contest between low-cost Chinese open models and premium closed American systems.

Jul 16, 2026 · Moonshot AI

Kimi K3 takes the top coding spot — at a fraction of the price

A 2.8-trillion-parameter open-weight model from Beijing edges past Anthropic's Claude Fable 5 on the Frontend Code Arena, sharpening the contest between low-cost Chinese open models and premium closed U.S. systems.

1,679

Frontend Code Arena score — #1 ranking

2.8T

Parameters — only 16 of 896 experts active per token

1.05M

Token context window, with native vision

6/7

Measured domains led on the Arena

Output price per 1M tokens

Kimi K3 undercuts Fable 5 by more than 3× on output — each block = $5.

$15

Kimi K3

$50

Claude Fable 5

Input: ~$3 (cache ~$0.30) vs ~$10 per 1M tokens.

Terminal Bench 2.1

A close race — the two models trade wins rather than one dominating.

88.3

Kimi K3

84.6

Fable 5

lead

GPT-5.6 Sol

The draw

Cleaner frontend & UI generation at a fraction of the cost
Strong on long-horizon software and game-dev prototypes
Open weights due Jul 27 — fine-tuning & local deployment

The caveats

Can consume more tokens and run slower
Fable 5 stays steadier on complex multi-step agentic work
Vendor benchmarks depend on the test environment

Not a single winner — a widening menu.

Cost-performance and openness increasingly rival raw benchmark scores as decision factors, even as export controls, distillation questions, and compliance debates keep circulating.

Continue reading

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