Benchmarks / Math

MGSM (Vals)

A multilingual benchmark for mathematical questions. Scores are the values published by VALS AI. Source benchmark version 1.

61models
–Open Index weight
2026-01-09last updated
Vals AIpublisher

MGSM (Vals)

Higher is better · best effort per model

Chart settings
61 of 61 selected

full table below

#ModelEffortScoreCost / taskHarness
1Opus 4.5high95.2%$0.063–
2Opus 4.1high94.4%$0.055–
3Sonnet 4.5high94.3%$0.011–
4GPT-5.2xhigh94.0%$0.011–
5Gemini 3 Prohigh93.9%$0.018–
6Opus 4medium93.8%$0.027–
7o4 minihigh93.4%$0.0029–
8Gemini 3 Flashhigh93.3%$0.0071–
9Sonnet 4high93.0%$0.0058–
10GPT-5.1high93.0%$0.0061–
11Sonnet 3.7medium93.0%$0.026–
12GPT-5high92.8%$0.014–
13Claude 3.5 Sonnet Latest–92.6%$0.0040–
14GPT-5 Minihigh92.6%$0.0037–
15Qwen 3 235B–92.5%$0.0011–
16Llama 4 Maverickmedium92.4%$0.0002–
17DeepSeek R1xhigh92.3%$0.0041–
18Haiku 4.5xhigh92.1%$0.0069–
19DeepSeek V3none92.1%$0.0004–
20Qwen 3 Maxmax92.1%$0.0023–
21gpt-oss-120bhigh92.0%$0.0007–
22o3high91.7%$0.0038–
23Grok 3none91.3%$0.0057–
24o3 minihigh91.3%$0.0082–
25Llama 3.3none91.1%$0.0005–
26Kimi K2medium90.9%$0.0010–
27Grok 4high90.9%$0.026–
28Grok 4 Fastnone90.9%$0.0006–
29Mistral Medium 3.1none90.9%$0.0005–
30DeepSeek V3.2high90.9%$0.0001–
31GLM 4.5xhigh90.8%$0.0063–
32GPT-4ohigh90.7%$0.0029–
33Grok 3 Minihigh90.4%$0.0008–
34Gemini 2.5 Flashmedium89.9%$0.0010–
35GLM 4.6none89.7%$0.0043–
36Gemini 2.5 Flash Litenone89.5%$0.0002–
37Grok 4.1 Fasthigh89.5%$0.0005–
38o1high89.3%$0.057–
39GPT-5 Nanohigh89.3%$0.0019–
40Gemini 1.5 Pronone89.2%$0.0010–
41Gemini 2.0 Flashmedium89.0%$0.0001–
42gpt-oss-20bhigh89.0%$0.0005–
43GLM 4.7none88.2%$0.0043–
44Llama 4 Scoutmedium88.0%$0.0001–
45MiniMax M2.1–87.9%$0.0013–
46GPT-4.1 Minihigh87.8%$0.0004–
47GPT-4.1high87.7%$0.0017–
48Mistral Large 11–87.2%$0.0030–
49Gemini 1.5 Flashnone86.6%$0.0001–
50Magistral Small 1.2–86.3%$0.0015–
51GPT-4o minihigh86.2%$0.0002–
52Grok 2none86.1%$0.0038–
53Command Anone85.7%$0.0037–
54Mistral Large 3none85.4%$0.0005–
55Haiku 3.5medium84.6%$0.0011–
56Mistral Small 3.1none84.2%$0.0001–
57Mistral Small 02–84.0%$0.0002–
58Magistral Medium 1.2medium74.6%$0.0057–
59Jamba 1.6 Largenone71.2%$0.0038–
60GPT-4.1 Nanohigh69.3%$0.0001–
61Jamba 1.6 Mininone41.7%$0.0003–