Vals AI SAGE: leaderboard
Evaluates models as AI graders for student math work, applying rubrics to handwritten solutions in automata theory, linear algebra, PDEs, and AP Calculus.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around 2032. 91 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (Max) | 56.1 | #49 (Claude Opus 4.7) |
| 2 | Claude Opus 4.8 (Max) | 54.79 | #37 (Claude Opus 4.8) |
| 3 | Gemini 4 Argon (High) | 53.65 | #8 (Gemini 4 Argon) |
| 4 | Kimi K3 (Max) | 52.78 | #30 (Kimi K3) |
| 5 | GPT-5.6 Sol (Max) | 52.56 | #19 (GPT-5.6 Sol) |
| 6 | Qwen 3.8 27B (xHigh) | 52.4 | #105 (Qwen 3.8 27B) |
| 7 | Claude Opus 4.5 (20251101) (Thinking, High) | 52.09 | #74 (Claude Opus 4.5 (20251101)) |
| 8 | Claude Fable 5 (Max) | 51.89 | #16 (Claude Fable 5) |
| 9 | Gemini 3 Flash (Preview) (High) | 51.85 | #82 (Gemini 3 Flash (Preview)) |
| 10 | Claude Sonnet 5.5 (Max) | 51.77 | #11 (Claude Sonnet 5.5) |
| 11 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort) | 51.58 | #63 (Claude Opus 4.6) |
| 12 | GPT-5.5 (xHigh) | 51.53 | #31 (GPT-5.5) |
| 13 | Qwen 3.8 Max | 51.25 | #35 |
| 14 | GPT-5.4 Mini (2026-03-17) (xHigh) | 50.81 | #169 (GPT-5.4 Mini (2026-03-17)) |
| 15 | MiniMax-M3 | 50.57 | #136 |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-sage · How It Works · Data refreshed daily, snapshot 2026-10-20.