# Mathematical Reasoning on AIME 2025 (Mean@32)

Source: [Mathematical Reasoning on AIME 2025 (Mean@32)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-aime-2025-mean-at-32>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-08T07:49:10.651Z

Evaluation of model performance on the American Invitational Mathematics Examination (AIME) 2025 dataset, reporting the Mean score calculated over 32 samples per problem.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Mathematical Reasoning |
| Dataset (including any stated split) | AIME 2025 |
| Metric | Mean Score (32 Samples) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Mean Score (32 Samples)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-aime-2025-mean-at-32.md?metric=Mean%20Score%20(32%20Samples)>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Mean Score (32 Samples) | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen2.5-Math-7B + Resample w/ LOPE (w/ Training Signal Shaping) | 16.51 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-4B-Base + Resample w/ LOPE (w/ Training Signal Shaping) | 16.27 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-4B-Base + Resample w/ LOPE (w/o Training Signal Shaping) | 13.85 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-4B-Base + GRPO | 13.12 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-4B-Base + Resample w/ Naive Prompt | 11.67 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen2.5-Math-7B + Resample w/ LOPE (w/o Training Signal Shaping) | 10.52 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen2.5-Math-7B + GRPO | 9.9 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen2.5-Math-7B + Resample w/ Naive Prompt | 9.17 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen2.5-Math-7B | 7.9 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-4B-Base | 7.24 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-1.7B-Base + Resample w/ LOPE (w/o Training Signal Shaping) | 5.83 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-1.7B-Base + Resample w/ LOPE (w/ Training Signal Shaping) | 5.73 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-1.7B-Base + Resample w/ Naive Prompt | 4.58 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-1.7B-Base + GRPO | 4.47 | Not recorded | 69fd6df75497b3b670c89346 |
| Qwen3-1.7B-Base | 2 | Not recorded | 69fd6df75497b3b670c89346 |

### Qwen2.5-Math-7B + Resample w/ LOPE (w/ Training Signal Shaping)

Mean Score (32 Samples): 16.51. Result ID: 69fd957347f34ce924e05a62.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Math-7B","Optimization Method":"GRPO + Resample","Prompt Strategy":"LOPE","Training Signal Shaping":"true"}. This record does not assert independent reproduction.

### Qwen3-4B-Base + Resample w/ LOPE (w/ Training Signal Shaping)

Mean Score (32 Samples): 16.27. Result ID: 69fd957347f34ce924e05a5d.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B-Base","Optimization Method":"GRPO + Resample","Prompt Strategy":"LOPE","Training Signal Shaping":"true"}. This record does not assert independent reproduction.

### Qwen3-4B-Base + Resample w/ LOPE (w/o Training Signal Shaping)

Mean Score (32 Samples): 13.85. Result ID: 69fd957347f34ce924e05a5c.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B-Base","Optimization Method":"GRPO + Resample","Prompt Strategy":"LOPE","Training Signal Shaping":"false"}. This record does not assert independent reproduction.

### Qwen3-4B-Base + GRPO

Mean Score (32 Samples): 13.12. Result ID: 69fd957347f34ce924e05a5a.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Model":"Qwen3-4B-Base","Optimization Method":"GRPO"}. This record does not assert independent reproduction.

### Qwen3-4B-Base + Resample w/ Naive Prompt

Mean Score (32 Samples): 11.67. Result ID: 69fd957347f34ce924e05a5b.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B-Base","Optimization Method":"GRPO + Resample","Prompt Strategy":"Naive Prompt"}. This record does not assert independent reproduction.

### Qwen2.5-Math-7B + Resample w/ LOPE (w/o Training Signal Shaping)

Mean Score (32 Samples): 10.52. Result ID: 69fd957347f34ce924e05a61.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Math-7B","Optimization Method":"GRPO + Resample","Prompt Strategy":"LOPE","Training Signal Shaping":"false"}. This record does not assert independent reproduction.

### Qwen2.5-Math-7B + GRPO

Mean Score (32 Samples): 9.9. Result ID: 69fd957347f34ce924e05a5f.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Math-7B","Optimization Method":"GRPO"}. This record does not assert independent reproduction.

### Qwen2.5-Math-7B + Resample w/ Naive Prompt

Mean Score (32 Samples): 9.17. Result ID: 69fd957347f34ce924e05a60.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Math-7B","Optimization Method":"GRPO + Resample","Prompt Strategy":"Naive Prompt"}. This record does not assert independent reproduction.

### Qwen2.5-Math-7B

Mean Score (32 Samples): 7.9. Result ID: 69fd957347f34ce924e05a5e.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74d81e209131cea6419f.md>)

[Implementation](<https://github.com/QwenLM/Qwen2-Math>)

Recorded evaluation configuration: {"Model":"Qwen2.5-Math-7B","Optimization Method":"Base"}. This record does not assert independent reproduction.

### Qwen3-4B-Base

Mean Score (32 Samples): 7.24. Result ID: 69fd957347f34ce924e05a59.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B-Base","Optimization Method":"Base"}. This record does not assert independent reproduction.

### Qwen3-1.7B-Base + Resample w/ LOPE (w/o Training Signal Shaping)

Mean Score (32 Samples): 5.83. Result ID: 69fd957347f34ce924e05a57.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen3-1.7B-Base","Optimization Method":"GRPO + Resample","Prompt Strategy":"LOPE","Training Signal Shaping":"false"}. This record does not assert independent reproduction.

### Qwen3-1.7B-Base + Resample w/ LOPE (w/ Training Signal Shaping)

Mean Score (32 Samples): 5.73. Result ID: 69fd957347f34ce924e05a58.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen3-1.7B-Base","Optimization Method":"GRPO + Resample","Prompt Strategy":"LOPE","Training Signal Shaping":"true"}. This record does not assert independent reproduction.

### Qwen3-1.7B-Base + Resample w/ Naive Prompt

Mean Score (32 Samples): 4.58. Result ID: 69fd957347f34ce924e05a56.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

Recorded evaluation configuration: {"Model":"Qwen3-1.7B-Base","Optimization Method":"GRPO + Resample","Prompt Strategy":"Naive Prompt"}. This record does not assert independent reproduction.

### Qwen3-1.7B-Base + GRPO

Mean Score (32 Samples): 4.47. Result ID: 69fd957347f34ce924e05a55.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Model":"Qwen3-1.7B-Base","Optimization Method":"GRPO"}. This record does not assert independent reproduction.

### Qwen3-1.7B-Base

Mean Score (32 Samples): 2. Result ID: 69fd957347f34ce924e05a54.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6df75497b3b670c89346.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Model":"Qwen3-1.7B-Base","Optimization Method":"Base"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
