PulseAugur
LIVE 13:12:51
research · [1 source] ·
0
research

Anthropic's Opus 4.6 creates playable CLI versions of Slay the Spire and Balatro

A researcher tested Opus 4.6's ability to recreate simplified command-line versions of the video games Slay the Spire and Balatro. Despite expecting failure, the AI successfully generated mostly playable, albeit buggy, implementations of both games. The agent was given a large context window and internet access to complete the task, which involved core game mechanics rather than full features. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

RANK_REASON The item describes an experiment and observations on an AI model's capabilities, akin to a research paper.

Read on METR (Model Evaluation & Threat Research) →

Anthropic's Opus 4.6 creates playable CLI versions of Slay the Spire and Balatro

COVERAGE [1]

  1. METR (Model Evaluation & Threat Research) TIER_1 ·

    Observations from two CLI game reimplementation runs with Opus 4.6

    <p>Summary: Opus 4.6 can, with a simple agent scaffold, create mostly-playable but somewhat broken CLI versions of Slay the Spire and Balatro<sup id="fnref:1"><a class="footnote" href="#fn:1" rel="footnote">1</a></sup>.</p> <h2 id="intro">Intro</h2> <p>Last weekend I was trying t…