Breaking

Moonshot's Kimi K3 Tops OpenAI and Anthropic Flagships in Head-to-Head Game-Design Test at a Fraction of the Cost

July 18, 2026 at 22:15 EDT

Foundation Models
Software Dev & Coding
Media Generation

A side-by-side comparison of three frontier AI models building games from an identical prompt handed a clear win to Moonshot AI's newly released Kimi K3, which scored 9.5 out of 10 while costing dramatically less to run than OpenAI's GPT-5.6 Sol and Anthropic's Fable 5.

July 19, 2026 · Frontier Model Face-off

Kimi K3 wins the build-a-game test — at a fraction of the cost

Given an identical game-design brief, Moonshot AI's Kimi K3 topped OpenAI's GPT-5.6 Sol and Anthropic's Fable 5, scoring 9.5/10 while running dramatically cheaper per attempt.

9.5/10

Kimi K3 composite score — design, features & cost

12.6×

cheaper per run than Fable 5

3.7×

cheaper per run than GPT-5.6 Sol

Cost per run · same unit ($)

The pricing gap, drawn to scale

$0.030

Kimi K3

$0.11

GPT-5.6 Sol

$0.38

Fable 5

Column height ∝ dollars per run. Kimi K3's sliver barely registers against Fable 5's stack.

The scorecard

Ranked on design, features & cost

Kimi K3

9.5

Fable 5

7.5

GPT-5.6 Sol

7.0

Both Fable 5 and GPT-5.6 Sol drew the same knock: their generated gameplay felt too fast. Kimi K3 kept pacing natural and chose the right features.

Why it matters

2.8-trillion-parameter, natively multimodal, 1M-token context
API at $3 in / $15 out per M tokens — as low as $0.30 on cache hits
Weights slated to become openly available

The caveats

Inference speed can lag in some cases
A narrow, single-domain snapshot — not broad reasoning
Fable 5 & GPT-5.6 Sol still lead on complex agentic work

Continue reading

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