Breaking

Moonshot AI's Kimi K3 Tops Frontend Code Arena, Edging Out Claude Fable 5

July 17, 2026 at 18:10 EDT

Foundation Models
Software Dev & Coding
Open Source

Moonshot AI's newly released Kimi K3 has climbed to the top of the Frontend Code Arena leaderboard, scoring 1,679 points to overtake Claude Fable 5 and claim the No. 1 spot in a benchmark focused on AI-generated front-end code. The result marks a dramatic jump for the Chinese lab's Kimi series, which rose from No. 18 with the prior Kimi-k2.6 to first place with the latest model.

July 16, 2026 · Moonshot AI

Kimi K3 Takes No. 1 on the Frontend Code Arena

Moonshot AI's new open model scores 1,679 points, overtaking Claude Fable 5 to lead a benchmark for AI-generated front-end code — a leap from No. 18 to No. 1 for the Kimi series.

Leaderboard rank of the Kimi series

#18

Kimi-k2.6
(previous)

→

Kimi K3
(latest)

Domains won of 7 tracked

6/7

1st in Brand & Marketing, Reference Design, Data & Analytics, Consumer Product, Simulations, and Content Tools. 2nd in Gaming.

2.8T

parameters (large open model)

token context window, native vision

~1/3

of Fable 5's price, by some accounts

API pricing — per million tokens

Input

$15

Output

Output tokens cost 5× input — a factor in reported heavy token consumption.

Developer strengths

Beats Fable 5 on many front-end tasks
Long-horizon coding across large repos
Strong tool use, debugging & SVG output
Cheaper, with open weights on the way

Caveats

Heavy token consumption & cost
Rate-limit errors from capacity limits
Fable 5 still leads in Gaming
Full open-source status unconfirmed

Open weights due July 27 — moving from API-only toward a downloadable open-weight model.

Built on Kimi Delta Attention (KDA), a hybrid linear-attention mechanism with attention residuals for efficiency at scale — pushing open models into frontier-grade coding territory.

Continue reading

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