# MGSM leaderboard

Multilingual benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelchoice.dev/leaderboards/benchmarks/multilingual-mgsm

Multilingual Grade School Math. Every model the catalog carries a published MGSM value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Multilingual |
| Measure | Math word problems |
| Tasks | 250 problems × 11 languages |
| Difficulty | Grade school math, multilingual |
| Published by | [Language Models are Multilingual Chain-of-Thought Reasoners](https://arxiv.org/abs/2210.03057) |

A multilingual benchmark that translates 250 grade school math problems from GSM8K into 10 typologically diverse languages: Bengali, German, Spanish, French, Japanese, Russian, Swahili, Telugu, Thai, and Chinese.

## MGSM ranking

13 models with a published MGSM value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Math word problems | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [o4-mini](https://modelchoice.dev/models/o4-mini) | OpenAI | 93.7 | Observed 2026-11-21 · source benchlm:models |
| 2 | [Llama 3.3 70B](https://modelchoice.dev/models/llama-3-3-70b) | Meta | 91.1 | Observed 2026-11-21 · source benchlm:models |
| 3 | [o1-mini](https://modelchoice.dev/models/o1-mini) | OpenAI | 89.9 | Observed 2026-11-21 · source benchlm:models |
| 4 | [GPT-4.5](https://modelchoice.dev/models/gpt-4-5) | OpenAI | 86.9 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Llama 3.1 70B](https://modelchoice.dev/models/llama-3-1-70b) | Meta | 86.9 | Observed 2026-11-21 · source benchlm:models |
| 6 | [Claude 3.5 Haiku](https://modelchoice.dev/models/claude-3-5-haiku) | Anthropic | 85.6 | Observed 2026-11-21 · source benchlm:models |
| 7 | [Phi-4](https://modelchoice.dev/models/phi-4) | Microsoft | 80.6 | Observed 2026-11-21 · source benchlm:models |
| 8 | [Llama 3.1 8B](https://modelchoice.dev/models/llama-3-1-8b) | Meta | 68.9 | Observed 2026-11-21 · source benchlm:models |
| 9 | [Phi-3.5-mini](https://modelchoice.dev/models/phi-3-5-mini) | Microsoft | 47.9 | Observed 2026-11-21 · source benchlm:models |
| 10 | [Granite-4.0-1B](https://modelchoice.dev/models/granite-4-0-1b) | IBM | 37.84 | Observed 2026-11-21 · source benchlm:models |
| 11 | [Granite-4.0-H-1B](https://modelchoice.dev/models/granite-4-0-h-1b) | IBM | 27.52 | Observed 2026-11-21 · source benchlm:models |
| 12 | [Granite-4.0-H-350M](https://modelchoice.dev/models/granite-4-0-h-350m) | IBM | 16.16 | Observed 2026-11-21 · source benchlm:models |
| 13 | [Granite-4.0-350M](https://modelchoice.dev/models/granite-4-0-350m) | IBM | 14.72 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
