MGSM

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

Published by

Vals AI

Category

Maths

Index weight

0.5

Models

70

Data as of

9 Sept 2026

Vals AI (vals.ai).

1. 1Claude Opus 4.595.2%
2. 2Claude Opus 4.594.8%
3. 3Claude Opus 4.194.4%
4. 4Claude Sonnet 4.594.3%
5. 5Claude Opus 4.194.2%
6. 6GPT-5.294.0%
7. 7Gemini 3 Pro93.9%
8. 8Claude Opus 493.8%
9. 9o4-mini93.4%
10. 10Gemini 3 Flash93.3%
11. 11Claude Sonnet 493.0%
12. 12GPT-5.193.0%
13. 13Claude 3.7 Sonnet93.0%
14. 14GPT-592.8%
15. 15GPT-5 mini92.6%

SearchAll providersAI21 LabsAlibabaAnthropicCohereDeepSeekGoogleMetaMiniMaxMistral AIMoonshot AIOpenAIxAIZhipu AIOpen weights70 of 70

#	Model	Score ↓	Run ↕	Index ↕	Published ↕
1	Claude Opus 4.5Anthropic	95.2%	–	60.6	–
2	Claude Opus 4.5Anthropic	94.8%	–	58.0	–
3	Claude Opus 4.1Anthropic	94.4%	–	58.1	–
4	Claude Sonnet 4.5Anthropic	94.3%	–	52.8	–
5	Claude Opus 4.1Anthropic	94.2%	–	52.6	–
6	GPT-5.2OpenAI	94.0%	–	57.3	–
7	Gemini 3 ProGoogle	93.9%	–	57.0	–
8	Claude Opus 4Anthropic	93.8%	–	50.7	–
9	o4-miniOpenAI	93.4%	–	48.0	–
10	Gemini 3 FlashGoogle	93.3%	–	58.5	–
11	Claude Sonnet 4Anthropic	93.0%	–	49.5	–
12	GPT-5.1OpenAI	93.0%	–	53.8	–
13	Claude 3.7 SonnetAnthropic	93.0%	–	52.1	–
14	GPT-5OpenAI	92.8%	–	54.5	–
15	GPT-5 miniOpenAI	92.6%	–	52.3	–
16	Claude 3.5 SonnetAnthropic	92.6%	–	45.1	–
17	Qwen3 235B A22BAlibabaopen	92.5%	–	48.7	–
18	Llama 4 Maverick BasicMeta	92.4%	–	38.8	–
19	Claude 3.7 SonnetAnthropic	92.4%	–	49.8	–
20	DeepSeek R1DeepSeekopen	92.3%	fireworks/deepseek-r1	48.9	–
21	Qwen3 MaxAlibaba	92.2%	–	51.8	–
22	Claude Haiku 4.5Anthropic	92.2%	–	48.0	–
23	DeepSeek V3DeepSeekopen	92.2%	fireworks/deepseek-v3	43.8	–
24	gpt-oss-120bOpenAIopen	92.0%	–	47.6	–
25	Qwen3 MaxAlibaba	91.8%	–	52.7	–
26	o3OpenAI	91.8%	–	52.9	–
27	Grok 3xAI	91.3%	grok/grok-3	50.0	–
28	o3-miniOpenAI	91.3%	–	47.8	–
29	Llama Llama 3.3 70B TurboMeta	91.1%	–	38.6	–
30	Kimi K2Moonshot AIopen	91.0%	–	48.8	–
31	Grok 4xAI	90.9%	0709	57.2	–
32	Claude Sonnet 4Anthropic	90.9%	–	54.0	–
33	Grok 4 FastxAI	90.9%	–	53.0	–
34	Mistral Medium (latest)Mistral AI	90.9%	medium, 2505	45.8	–
35	GLM-4.5Zhipu AIopen	90.8%	zai/glm-4.5	50.2	–
36	GPT-4oOpenAI	90.7%	–	41.1	–
37	Grok 3 Mini FastxAI	90.4%	–	53.5	–
38	Grok 3 Mini FastxAI	90.4%	–	50.3	–
39	Kimi K2 ThinkingMoonshot AIopen	90.2%	–	52.8	–
40	Gemini 2.5 Flash 09Google	89.8%	preview, 2025	50.7	–
41	Gemini 2.5 Flash 09Google	89.8%	thinking, preview, 2025	52.5	–
42	GLM-4.6Zhipu AIopen	89.8%	zai/glm-4.6	47.6	–
43	Grok 4.1xAI	89.5%	–	55.9	–
44	Gemini 2.5 Flash Lite 09Google	89.5%	preview, 2025	45.7	–
45	GPT-5 nanoOpenAI	89.3%	–	45.9	–
46	o1OpenAI	89.3%	–	44.3	–
47	Gemini 1.5 Pro 002Google	89.2%	–	42.8	–
48	gpt-oss-20bOpenAIopen	89.0%	–	47.1	–
49	Gemini 2.0 FlashGoogle	89.0%	–	42.3	–
50	Gemini 2.5 Flash Lite 09Google	88.4%	–	48.8	–
51	GLM-4.7Zhipu AIopen	88.2%	zai/glm-4.7	54.9	–
52	Grok 4 FastxAI	88.0%	–	41.9	–
53	Llama Llama 4 Scout 17B 16eMeta	88.0%	–	35.8	–
54	MiniMax-M2MiniMaxopen	87.8%	minimax/MiniMax-M2.1	55.1	–
55	GPT-4.1 miniOpenAI	87.8%	–	45.9	–
56	GPT-4.1OpenAI	87.7%	–	46.2	–
57	Grok 4.1xAI	87.6%	–	47.9	–
58	Mistral Large 2Mistral AIopen	87.2%	2411	41.2	–
59	Gemini 1.5 Flash 002Google	86.6%	–	38.7	–
60	Magistral SmallMistral AIopen	86.3%	2509	40.5	–

Show 10 more