BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

MGSM

Grade-school maths in ten languages. Run by Vals AI.

As of 25 Oct 2026, Claude Opus 4.5 leads MGSM on BenchLeader with 95.2%, ahead of Claude Opus 4.5 at 94.8%, across 71 model configurations with a published result.

Published by
Vals AI
Category
Maths
Index weight
0.5
Models
71
Data as of
25 Oct 2026
Newest result
9 Jan 2026

The publisher has posted no new results on this board since 9 Jan 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.

Vals AI (vals.ai).

What the test looks like

MGSM translates grade-school maths word problems into ten languages, testing whether reasoning holds up outside English.

How it is scored

Accuracy averaged across languages, run by Vals AI.

What to keep in mind

The problems are easy for frontier models, so the score mostly reflects language coverage; it is saturated in English.

Top 15
  1. 1Claude Opus 4.5 (thinking)95.2%
  2. 2Claude Opus 4.594.8%
  3. 3Claude Opus 4.1 (thinking)94.4%
  4. 4Claude Sonnet 4.5 (thinking)94.3%
  5. 5Claude Opus 4.194.2%
  6. 6GPT-5.2 (xhigh)94.0%
  7. 7Gemini 3 Pro (high)93.9%
  8. 8Claude Opus 493.8%
  9. 9o4-mini (high)93.4%
  10. 10Gemini 3 Flash (high)93.3%
  11. 11Claude Sonnet 493.0%
  12. 12GPT-5.1 (high)93.0%
  13. 13Claude 3.7 Sonnet (thinking)93.0%
  14. 14GPT-5 (high)92.8%
  15. 15GPT-5 mini (high)92.6%
71 of 71
#
1Claude Opus 4.5thinking95.2%59.92026-01-09
2Claude Opus 4.594.8%57.92026-01-09
3Claude Opus 4.1thinking94.4%56.42026-01-09
4Claude Sonnet 4.5thinking94.3%52.62026-01-09
5Claude Opus 4.194.2%51.52026-01-09
6GPT-5.2xhigh94.0%60.72026-01-09
7Gemini 3 Prohigh93.9%60.02026-01-09
8Claude Opus 493.8%52.62026-01-09
9o4-minihigh93.4%50.12026-01-09
10Gemini 3 Flashhigh93.3%58.32026-01-09
11Claude Sonnet 493.0%50.52026-01-09
12GPT-5.1high93.0%57.72026-01-09
13Claude 3.7 Sonnetthinking93.0%51.02026-01-09
14GPT-5high92.8%57.92026-01-09
15GPT-5 minihigh92.6%54.12026-01-09
16Claude 3.5 Sonnet92.6%46.32026-01-09
17Qwen3 235B A22B92.5%50.12026-01-09
18Llama 4 Maverick Basic92.4%39.42026-01-09
19Claude 3.7 Sonnet92.4%50.02026-01-09
20DeepSeek R192.3%46.82026-01-09
21Qwen3 Maxmax92.2%52.82026-01-09
22Claude Haiku 4.5thinking92.2%46.72026-01-09
23DeepSeek V392.2%44.02026-01-09
24gpt-oss-120b92.0%45.62026-01-09
25o3high91.8%53.92026-01-09
26Grok 391.3%51.52026-01-09
27o3-minihigh91.3%47.22026-01-09
28Llama Llama 3.3 70B Turbo91.1%39.32026-01-09
29Kimi K291.0%50.42026-01-09
30Grok 490.9%58.02026-01-09
31Claude Sonnet 4thinking90.9%53.22026-01-09
32Grok 4 Fastthinking90.9%51.42026-01-09
33Mistral Medium 390.9%44.12026-01-09
34GLM-4.590.8%51.72026-01-09
35GPT-4ohigh90.7%41.72026-01-09
36Grok 3 Mini Fasthigh90.4%53.72026-01-09
37Grok 3 Mini Fastlow90.4%50.72026-01-09
38Kimi K2 Thinkingthinking90.2%52.42026-01-09
39Gemini 2.5 Flash 09 202589.8%46.92026-01-09
40Gemini 2.5 Flash 09 2025thinking89.8%51.12026-01-09
41GLM-4.689.8%49.92026-01-09
42Grok 4.1 Fastthinking89.5%54.72026-01-09
43Gemini 2.5 Flash Lite 0989.5%46.92026-01-09
44o1high89.3%48.32026-01-09
45GPT-5 nanohigh89.3%47.42026-01-09
46Gemini 1.5 Pro 00289.2%41.62026-01-09
47gpt-oss-20b89.0%44.12026-01-09
48Gemini 2.0 Flash89.0%43.22026-01-09
49Gemini 2.5 Flash Lite 09thinking88.4%47.92026-01-09
50GLM 4.788.2%51.92026-01-09
51Grok 4 Fastno reasoning88.0%44.22026-01-09
52Llama Llama 4 Scout 17B 16e88.0%36.52026-01-09
53MiniMax-M2.187.8%53.82026-01-09
54GPT-4.1 minihigh87.8%46.32026-01-09
55GPT-4.1high87.7%46.72026-01-09
56Grok 4.1 Fastno reasoning87.6%42.52026-01-09
57Mistral Large 2.187.2%42.52026-01-09
58Gemini 1.5 Flash 00286.6%38.32026-01-09
59Magistral Small 1.286.3%40.32026-01-09
60GPT-4o minihigh86.2%33.72026-01-09

More maths benchmarks · All LLM benchmarks

Cite as: BenchLeader, “MGSM leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/vals_mgsm.

MGSM: questions

What does MGSM measure?
MGSM translates grade-school maths word problems into ten languages, testing whether reasoning holds up outside English. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads MGSM?
Claude Opus 4.5 leads MGSM with 95.2% as of 25 Oct 2026, ahead of Claude Opus 4.5 at 94.8%.
How many models have MGSM results?
71 model configurations have a MGSM result on BenchLeader, all taken from Vals AI.
Who runs MGSM and how often is it updated?
MGSM is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
Does MGSM count toward the BenchLeader Index?
Yes. MGSM contributes to the maths category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.