Breaking

Moonshot AI's Kimi K3 Builds Playable Game in Three Prompts for $3.24, Undercutting Rival Models

July 18, 2026 at 00:15 EDT

Foundation Models
Software Dev & Coding
Open Source

Moonshot AI's newly released Kimi K3 generated a playable game blending Counter-Strike-style shooting with Portal-style spatial mechanics in just three prompt iterations, at an API cost of $3.24 — a fraction of what leading rivals would charge for the same workload. The demonstration, which circulated widely alongside gameplay footage, has become a focal point for developers weighing the practical economics of frontier coding models.

July 16, 2026 · Moonshot AI

Kimi K3 builds a CS:GO-meets-Portal game for $3.24

Moonshot AI's open-weight flagship generated a playable shooter with portal-style spatial mechanics in just three prompt iterations — at a fraction of what leading rivals would charge for the same ~600,000-token workload.

prompt iterations to a playable game

~600K

tokens consumed for the build

2.8T

parameters · 1M-token context

Same 600K-token job, three models

Estimated API cost to produce the CS:GO-meets-Portal clone

$3.24

Kimi K3

$6.00

GPT-5.6 Sol

$10.80

Fable 5

Fable 5 costs 3.3× more than Kimi K3 for the identical iterative build.

Under the hood

Architecture

Kimi Delta Attention + Stable LatentMoE — activates 16 of 896 experts per pass.

Pricing

$3 / 1M input · $15 / 1M output, with further cache-hit reductions.

Open weights

Full model weights slated for public release on July 27, 2026.

Where it shines

Top-tier on the Front-end Code Arena
Strong on long-running coding & agentic tasks
Native visual feedback — screenshots fed back across iterations
Standout cost-performance ratio

The caveats

Slightly behind Fable 5 & GPT-5.6 Sol on output polish
Price advantage not yet matched by a quality lead
Near — but not ahead — on FrontierSWE benchmarks
May signal the end of ultra-cheap Chinese AI

A Chinese open-weight model closing the gap with the frontier.

With competitive pricing already in market and weights due for public release, Kimi K3 sharpens the pressure on incumbents to justify higher token costs — targeting demanding work from GPU compiler and chip design to research reproduction, not one-off demos.

Continue reading

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