Add a benchmark

Benchmarks / Reasoning

SAGE

Student Assessment with Generative Evaluation: accuracy grading student math work against established rubrics.

80models
–Open Index weight
2026-10-08last updated
Vals AIpublisher

SAGE

Higher is better · best effort per model

Chart settings
80 of 80 selected

full table below

#ModelEffortScoreCost / taskHarness
1Opus 4.7max56.1%$0.331–
2Gemma 4 31B IThigh55.0%$0.0000–
3Opus 4.8max54.8%$0.505–
4Gemini 4high53.6%$0.098–
5Kimi K3max52.8%$0.163–
6GPT-5.6 Solmax52.6%$0.148–
7Qwen 3.8 27Bxhigh52.4%$0.036–
8Opus 4.5high52.1%$0.154–
9Fable 5max51.9%$0.652–
10Gemini 3 Flashhigh51.8%$0.018–
11Sonnet 5.5max51.8%$0.179–
12Opus 4.6max51.6%$0.353–
13GPT-5.5xhigh51.5%$0.136–
14Qwen 3.8 Maxmax51.3%$0.074–
15GPT-5.4 Minixhigh50.8%$0.076–
16MiniMax M3max50.6%$0.026–
17Kimi K2.6max50.2%$0.068–
18Gemini 3.5 Flashhigh49.9%$0.053–
19Kimi K2.5max49.9%$0.029–
20Gemini 3.1 Flash-Litehigh49.5%$0.0074–
21Opus 5max49.4%$0.193–
22GPT-5.2xhigh49.3%$0.116–
23Gemini 3.7 Flashhigh49.2%$0.026–
24Sonnet 5max48.9%$0.589–
25Gemini 3.6 Flashhigh48.7%$0.045–
26Gemini 3.1 Prohigh48.7%$0.069–
27Fable 5.1max48.5%$0.610–
28GPT-6 Lunamax48.1%$0.0065–
29DeepSeek V4.1 Flashhigh47.9%$0.012–
30Muse Spark 1.2xhigh47.7%$0.032–
31Gemini 3 Prohigh47.6%$0.012–
32Gemini 3.5 Flash-Litehigh47.3%$0.011–
33GPT-5.6 Terramax47.0%$0.038–
34Sonnet 4.6max46.6%$0.235–
35GPT-6.1 Solmax46.5%$0.061–
36GPT-6 Astramax46.4%$0.283–
37Opus 5.5max45.8%$0.332–
38Haiku 5.5max45.8%$0.019–
39Muse Spark 1.1xhigh45.6%$0.026–
40Qwen 3.6 27Bmax45.6%$0.050–
41MiMo V2.6 Prohigh45.0%$0.017–
42Qwen 3.6 Plusmax44.9%$0.041–
43GPT-6 Solmax44.8%$0.061–
44Gemini 2.5 Flashmedium44.8%$0.0027–
45GPT-5.6 Lunamax44.2%$0.011–
46GPT-5high43.7%$0.032–
47MiMo V2.6 Flashhigh43.5%$0.0049–
48GPT-5.4xhigh43.3%$0.098–
49MiMo V2.5xhigh43.3%$0.0044–
50GPT-5.1high43.2%$0.0048–
51GPT-5 Minihigh43.0%$0.0052–
52Qwen 3.5 Flashmax42.5%$0.0059–
53Qwen 3high42.1%$0.0048–
54Gemini 2.5 Prohigh41.9%$0.0088–
55o3high41.8%$0.028–
56o4 minihigh41.1%$0.029–
57Grok 4.7xhigh40.8%$0.325–
58Qwen 3.7 Plushigh39.3%$0.022–
59Grok 4.20xhigh38.2%$0.063–
60GPT-5.4 Nanohigh38.1%$0.0059–
61Mistral Medium 3.5high37.6%$0.073–
62Inklingmax36.6%$0.163–
63Sonnet 4.5high36.1%$0.138–
64Llama 4 Maverickmedium35.7%$0.0047–
65Gemini 3.8 Flashhigh35.1%$0.0090–
66Sonnet 4high35.0%$0.046–
67Grok 4.5high35.0%$0.116–
68Llama 4 Scoutmedium34.8%$0.0036–
69Inkling Smallxhigh32.7%$0.023–
70Haiku 4.5xhigh31.8%$0.053–
71Opus 4.1high31.4%$0.219–
72Grok 4.1 Fasthigh31.3%$0.0015–
73Gemini 2.5 Flash Litenone30.8%$0.0009–
74Qwen 3.5 Plusnone30.4%$0.038–
75GPT-5 Nanohigh30.4%$0.0023–
76Grok 4 Fastnone29.8%$0.012–
77Grok 4.6high28.9%$0.110–
78Grok 4high25.1%$0.405–
79Mistral Large 3none24.6%$0.0081–
80Grok 4.3xhigh19.7%$0.038–