Benchmarks · reasoning
SAGE
Student Assessment with Generative Evaluation: accuracy grading student math work against established rubrics.
22 models · 22 results · updated 2026-10-27 · vals.ai ↗
Leaderboard
Opus 5.5 max
More from this benchmark
1 published measuresTime per task28 results · lower is better
All results
22 rows| # | Model | Effort | SAGE | Subscription $ / task | API $ / task | Time |
|---|---|---|---|---|---|---|
| 1 | Opus 5.5 | max | 91.3% | $0.004 | $0.28 | 88s |
| 2 | GPT-6.1 Sol | max | 90.5% | $0.010 | $0.083 | 50s |
| 3 | Haiku 5.5 | max | 89.8% | $0.000 | $0.016 | 1.9 min |
| 4 | Gemini 4 Argon | high | 89.5% | $0.027 | $0.075 | 29s |
| 5 | GLM 5.3 Flash | max | 89.3% | free | $0.006 | 3.3 min |
| 6 | Qwen 3.7 Plus | high | 89.1% | $0.032 | $0.020 | 1.8 min |
| 7 | Grok 4.7 | xhigh | 88.7% | $0.010 | $0.11 | 2.0 min |
| 8 | Gemini 3.8 Flash | high | 88.2% | $0.003 | $0.076 | 67s |
| 9 | Sonnet 5.5 | max | 88.0% | $0.002 | $0.16 | 81s |
| 10 | GPT-5.6 Terra | max | 87.0% | $0.006 | $0.11 | 43s |
| 11 | Kimi K3 | max | 86.8% | $0.018 | $0.23 | 2.4 min |
| 12 | MiniMax M3 | max | 85.5% | $0.002 | $0.015 | 52s |
| 13 | Qwen 3.5 397B-A17B | max | 85.5% | $0.013 | $0.013 | 2.8 min |
| 14 | DeepSeek V4.1 Flash | max | 85.4% | $0.001 | $0.011 | 33s |
| 15 | Qwen 3.5 Flash | max | 84.9% | $0.006 | $0.006 | 1.7 min |
| 16 | MiMo V2.6 Pro | high | 84.4% | $0.018 | $0.021 | 2.4 min |
| 17 | GPT-6 Luna | max | 83.4% | $0.001 | $0.006 | 48s |
| 18 | MiMo V2.6 Flash | high | 83.3% | free | $0.007 | 2.8 min |
| 19 | Gemma 4 31B IT | high | 82.8% | free | 1.8 min | |
| 20 | Qwen 3.8 Max | max | 81.8% | $0.087 | $0.087 | 1.9 min |
| 21 | Gemini 3.5 Flash-Lite | high | 79.5% | $0.010 | $0.010 | 9.8s |
| 22 | Mistral Medium 3.5 | high | 72.0% | $0.010 | $0.074 | 47s |