# Vals MGSM (MGSM)

> A multilingual benchmark for mathematical questions.

Canonical page: https://benchlm.ai/benchmarks/valsmgsm

- Category: [Mathematics](/math)
- Last updated: January 9, 2026

## About MGSM

- Year: 2026
- Tasks: Multilingual grade-school math questions
- Format: Accuracy score
- Difficulty: Multilingual mathematical reasoning
- Paper: [MGSM](https://www.vals.ai/benchmarks/mgsm)

BenchLM mirrors the public Vals AI MGSM leaderboard as display-only external evidence. The captured snapshot preserves overall scores, task-level scores where Vals publishes them, uncertainty, latency, and cost-per-test metadata. It is excluded from BenchLM weighted rankings.

MGSM is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (75 models)

| Rank | Model | Configuration | Creator | Score |
|------|-------|---------------|---------|-------|
| 1 | [Claude Opus 4.5 Thinking](/models/claude-opus-4-5-thinking) | — | Anthropic | 95.20% |
| 2 | [Claude Opus 4.5](/models/claude-opus-4-5) | — | Anthropic | 94.76% |
| 3 | [Claude Opus 4.1 20250805 Thinking](https://www.vals.ai/models/anthropic_claude-opus-4-1-20250805-thinking) | — | Anthropic | 94.44% |
| 4 | [Claude Sonnet 4.5 Thinking](/models/claude-sonnet-4-5-thinking) | — | Anthropic | 94.33% |
| 5 | [Claude Opus 4.1](https://www.vals.ai/models/anthropic_claude-opus-4-1-20250805) | — | Anthropic | 94.22% |
| 6 | [GPT-5.2](/models/gpt-5-2) | xhigh reasoning | OpenAI | 94.00% |
| 7 | [Gemini 3 Pro Preview](https://www.vals.ai/models/google_gemini-3-pro-preview) | high reasoning | Google | 93.93% |
| 8 | [Claude Opus 4](https://www.vals.ai/models/anthropic_claude-opus-4-20250514) | — | Anthropic | 93.78% |
| 9 | [O4 Mini](https://www.vals.ai/models/openai_o4-mini-2025-04-16) | high reasoning | OpenAI | 93.42% |
| 10 | [Gemini 3 Flash Preview](https://www.vals.ai/models/google_gemini-3-flash-preview) | high reasoning | Google | 93.31% |
| 11 | [Claude Sonnet 4](https://www.vals.ai/models/anthropic_claude-sonnet-4-20250514) | — | Anthropic | 93.02% |
| 12 | [GPT-5.1](/models/gpt-5-1) | high reasoning | OpenAI | 92.98% |
| 13 | [Claude 3.7 Sonnet 20250219 Thinking](https://www.vals.ai/models/anthropic_claude-3-7-sonnet-20250219-thinking) | — | Anthropic | 92.98% |
| 14 | [GPT-5](https://www.vals.ai/models/openai_gpt-5-2025-08-07) | high reasoning | OpenAI | 92.84% |
| 15 | [Claude 3.5 Sonnet](/models/claude-3-5-sonnet) | — | Anthropic | 92.58% |
| 16 | [GPT-5 mini](/models/gpt-5-mini) | high reasoning | OpenAI | 92.58% |
| 17 | [Qwen3 235b A22b](https://www.vals.ai/models/fireworks_qwen3-235b-a22b) | — | Fireworks AI | 92.47% |
| 18 | [Llama4 Maverick Instruct Basic](https://www.vals.ai/models/fireworks_llama4-maverick-instruct-basic) | — | Fireworks | 92.44% |
| 19 | [Claude 3.7 Sonnet](https://www.vals.ai/models/anthropic_claude-3-7-sonnet-20250219) | — | Anthropic | 92.40% |
| 20 | [DeepSeek-R1](/models/deepseek-r1) | — | DeepSeek | 92.25% |
| 21 | [Claude Haiku 4.5 Thinking](/models/claude-haiku-4-5-thinking) | — | Anthropic | 92.15% |
| 22 | [DeepSeek V3](/models/deepseek-v3) | — | DeepSeek | 92.15% |
| 23 | [Qwen3 Max Preview](https://www.vals.ai/models/alibaba_qwen3-max-preview) | — | Alibaba | 92.15% |
| 24 | [GPT-OSS 120B](/models/gpt-oss-120b) | — | OpenAI | 92.04% |
| 25 | [Qwen3 Max](/models/qwen3-max) | — | Alibaba | 91.82% |
| 26 | [o3](/models/o3) | high reasoning | OpenAI | 91.75% |
| 27 | [DeepSeek V3 0324](https://www.vals.ai/models/fireworks_deepseek-v3-0324) | — | Fireworks AI | 91.67% |
| 28 | [Grok 3](https://www.vals.ai/models/grok_grok-3) | — | xAI | 91.35% |
| 29 | [o3-mini](/models/o3-mini) | high reasoning | OpenAI | 91.35% |
| 30 | [Meta Llama Llama 3.3 70B Instruct Turbo](https://www.vals.ai/models/together_meta-llama/Llama-3.3-70B-Instruct-Turbo) | — | Together AI | 91.09% |
| 31 | [Moonshotai Kimi K2 Instruct](https://www.vals.ai/models/together_moonshotai/Kimi-K2-Instruct) | — | Together AI | 90.95% |
| 32 | [Grok 4 0709](https://www.vals.ai/models/grok_grok-4-0709) | — | xAI | 90.91% |
| 33 | [Grok 4 Fast (Reasoning)](/models/grok-4-fast-reasoning) | — | xAI | 90.87% |
| 34 | [Mistral Medium 2505](https://www.vals.ai/models/mistralai_mistral-medium-2505) | — | Mistral AI | 90.87% |
| 35 | [DeepSeek V3p2](https://www.vals.ai/models/fireworks_deepseek-v3p2) | — | Fireworks | 90.87% |
| 36 | [Claude Sonnet 4 20250514 Thinking](https://www.vals.ai/models/anthropic_claude-sonnet-4-20250514-thinking) | — | Anthropic | 90.87% |
| 37 | [GLM-4.5](/models/glm-4-5) | — | Z.AI | 90.84% |
| 38 | [GPT-4o](/models/gpt-4o) | high reasoning | OpenAI | 90.69% |
| 39 | [Grok 3 Mini Fast High Reasoning](https://www.vals.ai/models/grok_grok-3-mini-fast-high-reasoning) | high reasoning | xAI | 90.44% |
| 40 | [GPT-4o](/models/gpt-4o) | high reasoning | OpenAI | 90.36% |
| 41 | [Grok 3 Mini Fast Low Reasoning](https://www.vals.ai/models/grok_grok-3-mini-fast-low-reasoning) | low reasoning | xAI | 90.36% |
| 42 | [Kimi K2 Thinking](https://www.vals.ai/models/kimi_kimi-k2-thinking) | — | Moonshot AI | 90.15% |
| 43 | [Gemini 2.5 Flash Preview 09 2025](https://www.vals.ai/models/google_gemini-2.5-flash-preview-09-2025) | — | Google | 89.85% |
| 44 | [Gemini 2.5 Flash Preview 09 2025 Thinking](https://www.vals.ai/models/google_gemini-2.5-flash-preview-09-2025-thinking) | — | Google | 89.82% |
| 45 | [GLM-4.6](/models/glm-4-6) | — | Z.AI | 89.75% |
| 46 | [Gemini 2.5 Flash Lite Preview 09 2025](https://www.vals.ai/models/google_gemini-2.5-flash-lite-preview-09-2025) | — | Google | 89.53% |
| 47 | [Grok 4.1 Fast (Reasoning)](/models/grok-4-1-fast-reasoning) | — | xAI | 89.53% |
| 48 | [o1](/models/o1) | high reasoning | OpenAI | 89.31% |
| 49 | [GPT-5 nano](/models/gpt-5-nano) | high reasoning | OpenAI | 89.31% |
| 50 | [Gemini 1.5 Pro 002](https://www.vals.ai/models/google_gemini-1.5-pro-002) | — | Google | 89.20% |
| 51 | [Gemini 2.0 Flash 001](https://www.vals.ai/models/google_gemini-2.0-flash-001) | — | Google | 89.02% |
| 52 | [GPT-OSS 20B](/models/gpt-oss-20b) | — | OpenAI | 89.02% |
| 53 | [Gemini 2.5 Flash Lite Preview 09 2025 Thinking](https://www.vals.ai/models/google_gemini-2.5-flash-lite-preview-09-2025-thinking) | — | Google | 88.40% |
| 54 | [GLM-4.7](/models/glm-4-7) | — | Z.AI | 88.18% |
| 55 | [Grok 4 Fast Non Reasoning](https://www.vals.ai/models/grok_grok-4-fast-non-reasoning) | — | xAI | 88.00% |
| 56 | [Meta Llama Llama 4 Scout 17B 16E Instruct](https://www.vals.ai/models/together_meta-llama/Llama-4-Scout-17B-16E-Instruct) | — | Together AI | 87.96% |
| 57 | [MiniMax M2.1](https://www.vals.ai/models/minimax_MiniMax-M2.1) | — | MiniMax | 87.85% |
| 58 | [GPT-4.1 mini](/models/gpt-4-1-mini) | high reasoning | OpenAI | 87.78% |
| 59 | [GPT-4.1](/models/gpt-4-1) | high reasoning | OpenAI | 87.67% |
| 60 | [Grok 4.1 Fast Non Reasoning](https://www.vals.ai/models/grok_grok-4-1-fast-non-reasoning) | — | xAI | 87.56% |
| 61 | [Mistral Large 2411](https://www.vals.ai/models/mistralai_mistral-large-2411) | — | Mistral AI | 87.24% |
| 62 | [Gemini 1.5 Flash 002](https://www.vals.ai/models/google_gemini-1.5-flash-002) | — | Google | 86.58% |
| 63 | [Magistral Small 2509](https://www.vals.ai/models/mistralai_magistral-small-2509) | — | Mistral AI | 86.25% |
| 64 | [GPT-4o mini](/models/gpt-4o-mini) | high reasoning | OpenAI | 86.18% |
| 65 | [Grok 2 1212](https://www.vals.ai/models/grok_grok-2-1212) | — | xAI | 86.15% |
| 66 | [DeepSeek V3p2 Thinking](https://www.vals.ai/models/fireworks_deepseek-v3p2-thinking) | — | Fireworks | 86.04% |
| 67 | [Command A 03 2025](https://www.vals.ai/models/cohere_command-a-03-2025) | — | Cohere | 85.71% |
| 68 | [Mistral Large 2512](https://www.vals.ai/models/mistralai_mistral-large-2512) | — | Mistral AI | 85.42% |
| 69 | [Claude 3.5 Haiku](https://www.vals.ai/models/anthropic_claude-3-5-haiku-20241022) | — | Anthropic | 84.62% |
| 70 | [Mistral Small 2503](https://www.vals.ai/models/mistralai_mistral-small-2503) | — | Mistral AI | 84.22% |
| 71 | [Mistral Small 2402](https://www.vals.ai/models/mistralai_mistral-small-2402) | — | Mistral AI | 83.96% |
| 72 | [Magistral Medium 2509](https://www.vals.ai/models/mistralai_magistral-medium-2509) | — | Mistral AI | 74.62% |
| 73 | [Jamba Large 1.6](https://www.vals.ai/models/ai21labs_jamba-large-1.6) | — | AI21 Labs | 71.24% |
| 74 | [GPT-4.1 nano](/models/gpt-4-1-nano) | high reasoning | OpenAI | 69.27% |
| 75 | [Jamba Mini 1.6](https://www.vals.ai/models/ai21labs_jamba-mini-1.6) | — | AI21 Labs | 41.71% |

## FAQ

### What does MGSM measure?

A multilingual benchmark for mathematical questions.

### Which model leads the published MGSM snapshot?

Claude Opus 4.5 Thinking currently leads the published MGSM snapshot with a score of 95.20%.

### How many models are evaluated on MGSM?

The January 9, 2026 contains 75 AI models.

### Does MGSM affect BenchLM's overall score?

Not directly. MGSM is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.