Breaking

Moonshot's 2.8-Trillion-Parameter Kimi K3 Tops Frontend Coding Leaderboard, Edging Out Anthropic's Claude Fable 5

July 17, 2026 at 16:46 EDT

Foundation Models
Software Dev & Coding
Open Source

China's Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight model that seized the top spot on a leading front-end coding leaderboard, narrowly beating Anthropic's Claude Fable 5 in blind developer voting. The launch, dated around July 16, 2026, is being positioned as the first open model at the 3-trillion-parameter scale.

July 16, 2026 · Moonshot AI

Kimi K3 Tops the Front-End Coding Leaderboard

A 2.8-trillion-parameter open-weight model — the first at the 3T scale — narrowly beats Anthropic's Claude Fable 5 in blind developer voting for web development, at a fraction of the price.

2.8T

total parameters — first open model at the 3-trillion scale

1.8%

of params active per pass — 16 of 896 MoE experts

token context window, with native vision support

Frontend Code Arena — blind vote Elo

Head-to-head developer votes for web-dev tasks (~1,757 votes)

1,679

Kimi K3

~1,672

Claude Fable 5

A narrow win on this leaderboard — but Fable 5 still leads on the broader benchmark set.

Benchmark scorecard vs Claude Fable 5

By Moonshot's own tally across 14 benchmarks

6 / 14

Kimi K3 wins

Terminal Bench 2.1 · Program Bench · SWE Marathon · Automation Bench · SpreadsheetBench 2 · BrowseComp

8 / 14

Fable 5 wins

Holds the advantage on software-engineering and several visual tasks, plus overall user experience.

Kimi API pricing — per million tokens

$0.30

input · cache hit

$3.00

input · cache miss

$15

output

Substantially cheaper than Fable 5 or GPT-5.6 Sol. Open weights arrive July 27, 2026, enabling self-hosting and fine-tuning.

The bullish read

Evidence Chinese labs are doing genuine model development, not just distilling U.S. systems. Strong UI generation and visual reasoning; open weights plus low cost lower the barrier to production deployment.

The cautious read

Moonshot admits K3 still trails Fable 5 and GPT-5.6 Sol on overall UX. Observers warn against over-reading one front-end leaderboard when Fable 5 leads on most broader benchmarks.

Continue reading

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