// tmb scoreboard
Which model, for which job.
Local and cloud models on the same real professional tasks, graded criterion by criterion.
- Models evaluated
- 88
- incl. 7 out of competition
- Cloud and local
- 45/43
- Tasks graded per model
- 32
- Judge
- Claude Opus 4.8
- 17 skills, 5 categories, 2026-09-27
Best for the job
Our recommendation per category, for the classes you selected. Models within 1 point of the leader are ties: at this level the gap is smaller than the judge's own noise. References and CoeOS are not recommended.
Quality against cost
Cloud Local Out of competition Best value frontier
Ranking
How it is scored
Real professional tasks, graded criterion by criterion against written grids. Claude Opus 4.8 scores each criterion, the harness does the arithmetic, and we calibrate the judge against others. References and CoeOS are shown out of competition.