Mun logadan · 2026-08-14 · notable
Mun logadan — benchmarks reward guessing, so Claude Opus 5 stops asking
Mun logadan argues that Claude Opus 5 feels worse to code with because benchmark training rewards models that guess confidently instead of asking what the developer meant. The post hit the Hacker News front page with 182 points.
One developer's theory for why a stronger model can be more annoying to work with: benchmarks punish asking questions.
What is it?
'Why does Opus 5 feel worse to work with?' takes on a complaint developers have repeated since the model shipped — Claude Opus 5 scores higher than its predecessors yet feels more frustrating in everyday coding. Mun logadan's answer is that the model acts on its assumptions instead of checking them first. The post is short and the author labels the reasoning speculation; it carries no experiments or benchmark data.
How does it work?
The argument points at two pressures inside AI labs: the push toward self-improving systems, and the incentive to win on benchmarks. Selecting for benchmark scores, the post says, 'inherently selects for models that make bold, usually-correct assumptions in the face of ambiguity. It penalizes models with a tendency to stop and ask for clarification.'
Why does it matter?
The piece lands on a real gap between how coding models are graded and how they are used: a benchmark task has one right answer, a working codebase often does not. As the author puts it, 'with real-life consequences on the line, I do not want an agent taking its best guess.' The Hacker News thread on the post drew 158 comments.
Who is it for?
developers using coding agents