Breaking

Terraria Clone Test Pits Anthropic's Claude Fable 5 Against OpenAI's GPT-5.6 Sol, With Cost and Quality Splitting the Verdict

July 13, 2026 at 00:45 EDT

Foundation Models
Software Dev & Coding
Media Generation

A single-prompt experiment that asked two of the industry's newest models to clone the sandbox game Terraria produced a telling split: Anthropic's Claude Fable 5 delivered the higher-quality result, while OpenAI's GPT-5.6 Sol reached a playable game at roughly a quarter of the cost and in about two-thirds of the time.

One-Shot Coding Showdown · Terraria Clone Test

Smarter Build vs. Cheaper Build: Two Models, One Prompt

Given the same maximum-effort instruction to clone Terraria, Claude Fable 5 produced the higher-quality game — while GPT-5.6 Sol reached a playable result at roughly a quarter of the cost and two-thirds of the time.

$120

Fable 5 cost · ~2 hours

$35

Sol cost · ~1h 20m

~3.4×

Cheaper with Sol

Cost of one Terraria clone

Height drawn to scale — each block ≈ $20.

$120

Claude Fable 5

$35

GPT-5.6 Sol

Benchmark split · TerminalBench 2.1

Sol Ultra

91.9%

Sol (standard)

88.8%

Claude Fable 5

84.3%

Claude Fable 5 — the smarter build

Stronger architecture & visual quality
Fewer iterations to a working result
~1M token context; edge on long tasks
Heavy token use; strict rate limits

GPT-5.6 Sol — the practical build

Wins on price and speed
Tops Design Arena · Elo 1353
~1.05M token context
$5 / $30 per M input / output tokens

The choice facing developers, prompt by prompt:

Pay more for the stronger build — or accept a slightly weaker result for a fraction of the cost. The Terraria clone is a snapshot of a market weighing capability against economics.

Continue reading

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