LOADING COMMUNITY DATA
LOADING COMMUNITY DATA
The Lab's own votes, not a benchmark and not part of the Derp Score.
Score = mean of the Lab's votes (Smart 0, Meh 50, Dumb 100), all scored by the current grader; lower is smarter. Per model only: nothing is added up across agents.
DAILY RUN
Every night each model does the same chores in real test projects, and the Lab votes on each: Smart 0, Meh 50, Dumb 100. Lower is better.
Newest run2026-11-17 18:30 UTC
Percent points of each plan's 5-hour and weekly windows. A model counts only chores measured while no other session of that CLI was working; the run total is first reading to last, on the owner's own accounts. Claude Code and Codex report whole percents.
Claude CodeMax
| Model | 5-hour | Weekly | Measured |
|---|---|---|---|
| Claude Fable 5.2 | +4.0%p | +1.0%p | 3/11 |
| Claude Opus 5.5 | +4.0%p | +1.0%p | 6/11 |
| Claude Sonnet 5.5 | +3.0%p | +1.0%p | 8/11 |
| Whole run | — | +6.0%p | |
| Used after the run | 51% | 56% |
OpenAI CodexPlus
| Model | 5-hour | Weekly | Measured |
|---|---|---|---|
| GPT-6.1 Sol | +23.0%p | +3.0%p | 11/11 |
| GPT-6.1 Terra | +14.0%p | +3.0%p | 11/11 |
| Whole run | +37.0%p | +6.0%p | |
| Used after the run | 37% | 27% |
Every run happens on one machine with the maintainer's own subscriptions, so time per chore depends on this setup and these plans' limits: compare runs here with each other, not with other setups.
| Model | 5-hour | Weekly | Measured |
|---|---|---|---|
| Gemini 4 Carbonhigh | +18.1%p | +4.9%p | 9/9 |
| Gemini 4 Argonhigh | +26.3%p | +8.5%p | 11/11 |
| Whole run | +44.4%p | +12.0%p | |
| Used after the run | 44% | 43% |