Benchmarks
348 sources · 1,258 models × 3,041 benchmarks · 158,957 scoresThe 258 GradeBench tasks where the student draws (graphs and plots, chemical structures, circuits and timing diagrams, orbital and energy-level diagrams, trees and graphs, free-body and ray diagrams), from 36 questions in 24 courses. Exact rubric match on those answers, from the same runs as the overall GradeBench score.
142nd strongest of 799 fitted benchmarks
Signal-to-noise +11.78 for the GradeBench family
Another benchmark represents this family in the fit.
| # | Model | Setting | Score | Source |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 | medium | 58.8 | |
| 2 | Gemini 4 Carbon | medium | 56.7 | |
| 3 | Gemini 4 Argon | medium | 55.5 | |
| 4 | GPT-6 Astra | medium | 54.3 | |
| 5 | GPT-6.1 Sol | medium | 52.9 | |
| 6 | Claude Opus 5.5 | medium | 51.4 | |
| 7 | Claude Fable 5.1 | medium | 51.2 | |
| 8 | Gemini 4 Hydrogen | medium | 51.0 | |
| 9 | Gemini 3.8 Flash | medium | 47.6 | |
| 10 | Muse Wonder | medium | 46.8 | |
| 11 | Muse Spark 1.3 | medium | 39.3 | |
| 12 | Claude Sonnet 5.5 | medium | 37.1 | |
| 13 | GPT-6 Luna | medium | 32.9 | |
| 14 | Grok 4.7 | medium | 29.6 | |
| 15 | GLM-5.3 Flash | not reported | 27.9 |
Benchmarks outside its family whose model rankings track it most reliably (Spearman ρ over at least 10 shared models).