← Multilingual benchmarks

MGSM leaderboard (Jan 2026): AI model scores

As of 9 Jan 2026, the highest reported score on MGSM is Claude Opus 4.5 (Thinking) (Anthropic) at 95.20%, per Vals AI. The source no longer updates this leaderboard.

Source: Vals AI leaderboard ↗ · Leaderboard dated 9 Jan 2026 · no longer updated by the source · Last checked 17 Nov 2026

MGSM

Maths questions in multiple languages.

Vals AI does not publish a cost per test for this benchmark, so there is no performance-vs-cost chart. Showing the ranking instead.

Highest reported score: Claude Opus 4.5 (Thinking) (Anthropic) at 95.20%.

Accuracy, highest first. Position is the source’s published rank.

  1. 1Claude Opus 4.5 (Thinking) Anthropic95.20%
  2. 2Claude Opus 4.5 (Nonthinking) Anthropic94.76%
  3. 3Claude Opus 4.1 (Thinking) Anthropic94.44%
  4. 4Claude Sonnet 4.5 (Thinking) Anthropic94.33%
  5. 5Claude Opus 4.1 (Nonthinking) Anthropic94.22%
  6. 6GPT 5.2 OpenAI94.00%
  7. 7Gemini 3 Pro (11/25) Google93.93%
  8. 8Claude Opus 4 (Nonthinking) Anthropic93.78%
  9. 9o4 Mini OpenAI93.42%
  10. 10Gemini 3 Flash (12/25) Google93.31%
  11. 11Claude Sonnet 4 (Nonthinking) Anthropic93.02%
  12. 12GPT 5.1 OpenAI92.98%
  13. 13Claude 3.7 Sonnet (Thinking) Anthropic92.98%
  14. 14GPT 5 OpenAI92.84%
  15. 15Claude 3.5 Sonnet Latest Anthropic92.58%
  16. 16GPT 5 Mini OpenAI92.58%
  17. 17Qwen 3 (235B) Alibaba92.47%
  18. 18Llama 4 Maverick Meta92.44%
  19. 19Claude 3.7 Sonnet (Nonthinking) Anthropic92.40%
  20. 20DeepSeek R1 DeepSeek92.25%
  21. 21Claude Haiku 4.5 (Thinking) Anthropic92.15%
  22. 22DeepSeek V3 DeepSeek92.15%
  23. 23Qwen 3 Max Preview Alibaba92.15%
  24. 24GPT OSS 120B OpenAI92.04%
  25. 25Qwen 3 Max Alibaba91.82%
MGSM: all published values (25 models). “–” means the source lists no value.
1Claude Opus 4.5 (Thinking)Anthropic95.20%
2Claude Opus 4.5 (Nonthinking)Anthropic94.76%
3Claude Opus 4.1 (Thinking)Anthropic94.44%
4Claude Sonnet 4.5 (Thinking)Anthropic94.33%
5Claude Opus 4.1 (Nonthinking)Anthropic94.22%
6GPT 5.2OpenAI94.00%
7Gemini 3 Pro (11/25)Google93.93%
8Claude Opus 4 (Nonthinking)Anthropic93.78%
9o4 MiniOpenAI93.42%
10Gemini 3 Flash (12/25)Google93.31%
11Claude Sonnet 4 (Nonthinking)Anthropic93.02%
12GPT 5.1OpenAI92.98%
13Claude 3.7 Sonnet (Thinking)Anthropic92.98%
14GPT 5OpenAI92.84%
15Claude 3.5 Sonnet LatestAnthropic92.58%
16GPT 5 MiniOpenAI92.58%
17Qwen 3 (235B)Alibaba92.47%
18Llama 4 MaverickMeta92.44%
19Claude 3.7 Sonnet (Nonthinking)Anthropic92.40%
20DeepSeek R1DeepSeek92.25%
21Claude Haiku 4.5 (Thinking)Anthropic92.15%
22DeepSeek V3DeepSeek92.15%
23Qwen 3 Max PreviewAlibaba92.15%
24GPT OSS 120BOpenAI92.04%
25Qwen 3 MaxAlibaba91.82%

Source: Vals AI · leaderboard dated 9 Jan 2026 · checked 17 Nov 2026 · archived (no longer updated)

Citation

MGSM, accuracy: Claude Opus 4.5 (Thinking) (Anthropic) had the top reported score at 95.20%. Original source: Vals AI, https://www.vals.ai/benchmarks/mgsm, leaderboard dated 9 Jan 2026; the source no longer updates this leaderboard. Via ChatJudge: https://chatjudge.ai/benchmarks/multilingual/mgsm, checked 17 Nov 2026.

Methodology and caveats

What it measures

Maths questions in multiple languages.

Grade-school maths word problems translated into 10 languages, testing reasoning outside English.

What the score means

Accuracy

Accuracy
Share of problems answered correctly, averaged across languages.

Run by Vals AI (benchmark by Google Research). Leaderboard dated 9 Jan 2026 (no longer updated by the source). Last checked 17 Nov 2026. Showing top 25 of 75 models listed.

Last updated by Vals in January 2026, so newer models are missing.

Source: Vals AI leaderboard ↗

Results from published AI leaderboards. Compare models within each benchmark; scores from different tests aren’t directly comparable. How we collect these results

Top model here: Claude Opus 4.5 (Thinking) (Anthropic). Try prompts for Claude →

Questions people ask

Which AI has the highest reported score on MGSM?

As of 9 Jan 2026, the highest reported score on MGSM is Claude Opus 4.5 (Thinking) (Anthropic) at 95.20%, per Vals AI. The source no longer updates this leaderboard.

What does MGSM measure?

Maths questions in multiple languages. It is run by Vals AI (benchmark by Google Research).

What does the MGSM score mean?

Accuracy. Share of problems answered correctly, averaged across languages.

How recent are these MGSM results?

MGSM: leaderboard dated 9 Jan 2026, no longer updated by the source; we last checked it on 17 Nov 2026.

Can I compare MGSM scores with other benchmarks?

Not directly. Each leaderboard uses its own tasks, settings and scoring, so compare models within one benchmark. ChatJudge does not run the tests; we copy the published scores with a link to the source.