DerpMeter Daily Coding Chores: Hydrogen Passes 9/10, Beats Argon and Carbon

DerpMeter's Daily Run is the only head-to-head so far that puts Hydrogen against its siblings Argon and Carbon on the same day and the same tasks: 7 models, 10 coding chores, run on November 12, 2026 at 18:30 UTC. The result: Hydrogen scored 25, ranking 5th — but within the Gemini family, it was the only model with no Dumb verdicts.

Method: 10 chores, how scoring works

Seven models ran the same 10 coding chores on the same day; each chore was judged Smart or Dumb, with lower scores better (Dumb verdicts cost heavily). The value of the test is the shared field: same day, same tasks, same scoring, so the differences between models are directly comparable. One caveat: this is a single sample — useful as a reference, not a stable ranking.

Full-field results

Inside the family: why Hydrogen won

Three things stand out. First, Hydrogen wasn't slightly faster — it was 3–4× faster: 20 minutes against Argon's 73 and Carbon's 82. Second, Hydrogen read it right the first time: only 1.0M tokens of context re-read, versus Argon's 9.6M and Carbon's 7.5M — Argon and Carbon kept re-reading context while Hydrogen got it in one pass. Third, Carbon "overthinks" small tasks: 5 chores judged Dumb, a 6/10 pass rate — consistent with the Carbon topic page's finding that simple tasks cost 3× more on Carbon. A flagship model is not necessarily the better choice for chores.

Across the field: mid-pack, not first class

But don't look only inside the family. Across the field, Hydrogen ranked 5th, with all four ahead of it rivals: Sonnet 5.5 at just 5 points, Sol and Terra at 15, Opus 5.5 at 20. That ordering matches the AA index: Hydrogen's xhigh 54 approaches Sonnet and Sol's 55, but on terminal work it still trails (Terminal-Bench 4.0: 54% vs 64% and 56%). Hydrogen is the steadiest Gemini for everyday coding chores, but across the field it remains mid-pack.