# PolyMath 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-poly-math

Every model the catalog carries a published PolyMath value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Multilingual |
| Measure | Cross-lingual mathematical reasoning |
| Tasks | Multilingual math problems |
| Difficulty | Advanced multilingual reasoning |
| Published by | [Qwen3.6 launch benchmarks](https://qwen.ai/blog?id=qwen3.6) |

A multilingual mathematical reasoning benchmark that tests whether math performance transfers across languages rather than only in English.

## PolyMath ranking

6 models with a published PolyMath 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 | Cross-lingual mathematical reasoning | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Qwen3.7 Max](https://modelchoice.dev/models/qwen3-7-max) | Alibaba | 86.5 | Observed 2026-11-21 · source benchlm:models |
| 2 | [Qwen3.7 Plus](https://modelchoice.dev/models/qwen3-7-plus) | Alibaba | 84.0 | Observed 2026-11-21 · source benchlm:models |
| 3 | [Claude Opus 4.5](https://modelchoice.dev/models/claude-opus-4-5) | Anthropic | 79.0 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Qwen3.5 397B](https://modelchoice.dev/models/qwen3-5-397b) | Alibaba | 73.3 | Observed 2026-11-21 · source benchlm:models |
| 5 | [K-EXAONE 2.0](https://modelchoice.dev/models/k-exaone-2-0) | LG AI Research | 71.3 | Observed 2026-11-21 · source benchlm:models |
| 6 | [Kimi K2.5](https://modelchoice.dev/models/kimi-k2-5) | Moonshot AI | 43.1 | 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.
