MMLU Pro measures broad academic reasoning with harder multi-step questions across STEM, humanities, and social sciences. It is used as a high-level signal for knowledge-intensive assistant quality.
Covers 14 consolidated subject groups for broad coverage across domains; 10-option multiple-choice format reduces random-guess advantage compared with 4-option exams.
Multiple-choice format can overstate capability when models eliminate distractors without full understanding; older public questions can leak into training corpora and inflate frontier scores.
Higher is better. Scores represent accuracy percentage. Top models score 70-85%.
Max score: 100
Based on ~12,032 samples, ~350 input tokens and ~50 output tokens per sample (400 total per sample).
Estimates from MMLU-Pro paper. Actual costs vary with reasoning effort and multi-turn prompting.
Multiple-choice format can overstate capability when models eliminate distractors without full understanding.
| # | Model | Score | Est. Eval Cost | Score/$ |
|---|---|---|---|---|
| 1 | 89.8 | $15.64 | 5.74 | |
| 2 | 89.5 | $15.64 | 5.72 | |
| 3 | 89.5 | $36.10 | 2.48 | |
| 4 | 89.0 | $3.91 | 22.8 | |
| 5 | 88.9 | $36.10 | 2.46 | |
| 6 | 88.2 | $3.91 | 22.6 | |
| 7 | 88.2 | $3.91 | 22.6 | |
| 8 | 88.0 | $108.29 | 0.81 | |
| 9 | 87.5 | $1.99 | 44.1 | |
| 10 | 87.5 | $21.66 | 4.04 | |
| 11 | 87.4 | $15.79 | 5.53 | |
| 12 | 87.3 | $108.29 | 0.81 | |
| 13 | 87.1 | $11.28 | 7.72 | |
| 14 | 87.0 | $11.28 | 7.71 | |
| 15 | 86.7 | $11.28 | 7.69 | |
| 16 | 86.6 | $21.66 | 4.00 | |
| 17 | 86.5 | $11.28 | 7.67 | |
| 18 | 86.3 | $0.0000 | — | |
| 19 | 86.2 | $11.28 | 7.64 | |
| 20 | 86.2 | $1.43 | 60.2 | |
| 21 | 86.0 | $21.66 | 3.97 | |
| 22 | 86.0 | $11.28 | 7.62 | |
| 23 | 86.0 | $11.28 | 7.62 | |
| 24 | 86.0 | $108.29 | 0.79 | |
| 25 | 86.0 | $21.66 | 3.97 | |
| 26 | 86.0 | $108.29 | 0.79 | |
| 27 | 85.9 | $15.79 | 5.44 | |
| 28 | 85.8 | $0.0000 | — | |
| 29 | 85.6 | $3.47 | 24.6 | |
| 30 | 85.4 | $0.0000 | — | |
| 31 | 85.4 | $0.0000 | — | |
| 32 | 85.3 | $13.24 | 6.44 | |
| 33 | 85.1 | $3.45 | 24.6 | |
| 34 | 85.1 | $8.54 | 9.97 | |
| 35 | 85.0 | $1.14 | 74.4 | |
| 36 | 85.0 | $1.43 | 59.4 | |
| 37 | 84.9 | $0.0000 | — | |
| 38 | 84.9 | $9.45 | 8.99 | |
| 39 | 84.8 | $4.03 | 21.0 | |
| 40 | 84.8 | $109.19 | 0.78 | |
| 41 | 84.4 | $10.83 | 7.79 | |
| 42 | 84.3 | $0.60 | 140 | |
| 43 | 84.3 | $2.35 | 35.8 | |
| 44 | 84.2 | $0.0000 | — | |
| 45 | Gemini 2.5 Flash Preview (Sep '25) (Reasoning) source artificial_analysis Evidence observed Nov 5, 2026 | 84.2 | $0.0000 | — |
| 46 | 84.1 | $99.26 | 0.85 | |
| 47 | 84.1 | $0.0000 | — | |
| 48 | 83.8 | $0.0000 | — | |
| 49 | 83.8 | $0.0000 | — | |
| 50 | 83.7 | $11.28 | 7.42 |