Breaking

Moonshot AI's Open-Weight Kimi K3 Renders a Mechanically Accurate Bugatti W16 Engine in Three.js

July 18, 2026 at 18:45 EDT

Foundation Models
Software Dev & Coding
Open Source

Moonshot AI's newly released Kimi K3 has generated a photorealistic, mechanically accurate Bugatti W16 engine in the browser using Three.js, drawing renewed attention to how far open-weight models have closed the gap with top proprietary systems on complex 3D coding tasks.

Moonshot AI · Kimi K3

An open model builds a running Bugatti W16 engine in the browser

Kimi K3 rendered a photorealistic, mechanically accurate W16 in Three.js — matching Anthropic's Claude Fable 5 on the same prompt, but burning far more tokens to get there.

2.8T

total parameters (MoE, 16 of 896 experts active per token)

token context window, with native vision

2.5×

more tokens used than Fable 5 to reach the finished render

Token cost of the same result

Both models finished the W16 engine — K3's iterative "Vision in the Loop" refinement consumed 2.5× the tokens.

1×

Claude Fable 5

2.5×

Kimi K3

Intelligence Index — near the frontier

An open model competitive with the strongest closed systems.

Kimi K3 (open)

Claude Fable 5

Vision in the Loop

K3 feeds its own rendered screenshots back in to correct and refine 3D and front-end output.

Generate code

→

Render screenshot

→

See & correct

The strength

Detailed, mechanically plausible 3D output — accurate internal geometry across engines, jets and open-world scenes rivaling the best closed models.

The trade-off

Verbose, iterative runs raise cost — output billed at $15 per million tokens. Some users await a more efficient successor before production use.

API pricing (per million tokens)

$0.30

input · cache hit

$3.00

input · otherwise

$15.00

output

Continue reading

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