Claude Code flunks Elden Ring test

The hot take is simple: benchmark wins and demos can still mask a large gap between impressive coding ability and robust general-purpose autonomy.

• –Anthropic markets Opus 4.6 as a strong agentic coding model, but this kind of anecdote shows how fragile current systems can be outside their comfort zone.
• –Elden Ring is a harsh test of perception, planning, and fast control loops, which exposes the limits of text-first agents more clearly than code benchmarks do.
• –The post is not a rigorous eval, but it is a useful signal of public skepticism around “AGI” claims that outpace everyday reliability.
• –For developers, the practical read is to treat Claude Code as a powerful assistant for bounded tasks, not as a drop-in general intelligence.