# Scientific Reasoning on SuperGPQA (Accuracy)

Source: [Scientific Reasoning on SuperGPQA (Accuracy)](<https://www.sota2.com/research/sota/scientific-reasoning-on-supergpqa-accuracy>)

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

Data updated: 2026-02-27T12:07:24.197Z

Evaluation of model performance on the SuperGPQA dataset for Scientific Reasoning tasks, measured using Accuracy.

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 | Scientific Reasoning |
| Dataset (including any stated split) | SuperGPQA |
| Metric | Accuracy |
| Unit | PERCENT |
| 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

- [Accuracy](<https://www.sota2.com/research/sota/scientific-reasoning-on-supergpqa-accuracy.md?metric=Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Kimi-K2 Base | 44.7 | Not recorded | 695de86653fdd5c3617439f3 |
| DeepSeek-V3.2 Exp Base | 43.6 | Not recorded | 695de86653fdd5c3617439f3 |
| DeepSeek-V3.1 Base | 42.3 | Not recorded | 695de86653fdd5c3617439f3 |
| MiMo-V2-Flash Base | 41.1 | Not recorded | 695de86653fdd5c3617439f3 |
| Jet-RL | 35.2 | Not recorded | 69705dede25ad2fa9d1ed833 |
| BF16 | 35.1 | Not recorded | 69705dede25ad2fa9d1ed833 |
| Initial | 31.8 | Not recorded | 69705dede25ad2fa9d1ed833 |
| BF16-Train-FP8-Rollout | 30.3 | Not recorded | 69705dede25ad2fa9d1ed833 |
| BF16 | 28.6 | Not recorded | 69705dede25ad2fa9d1ed833 |
| Jet-RL | 28.5 | Not recorded | 69705dede25ad2fa9d1ed833 |
| Initial | 25.5 | Not recorded | 69705dede25ad2fa9d1ed833 |
| Jet-RL | 19.9 | Not recorded | 69705dede25ad2fa9d1ed833 |
| BF16 | 15.9 | Not recorded | 69705dede25ad2fa9d1ed833 |
| BF16-Train-FP8-Rollout | 14.7 | Not recorded | 69705dede25ad2fa9d1ed833 |
| Initial | 12.4 | Not recorded | 69705dede25ad2fa9d1ed833 |

### Kimi-K2 Base

Accuracy: 44.7. Result ID: 6996a26a5698665ec4ebf133.

[Reporting paper](<https://www.sota2.com/research/paper/695de86653fdd5c3617439f3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a4a1e209131cea823a1.md>)

Recorded evaluation configuration: {"# Shots":"5-shot","# Activated Params":"32B","# Total Params":"1043B"}. This record does not assert independent reproduction.

### DeepSeek-V3.2 Exp Base

Accuracy: 43.6. Result ID: 6996a26a5698665ec4ebf135.

[Reporting paper](<https://www.sota2.com/research/paper/695de86653fdd5c3617439f3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/692fc410d6b0dd33429332f4.md>)

Recorded evaluation configuration: {"# Shots":"5-shot","# Activated Params":"37B","# Total Params":"671B"}. This record does not assert independent reproduction.

### DeepSeek-V3.1 Base

Accuracy: 42.3. Result ID: 6996a26a5698665ec4ebf134.

[Reporting paper](<https://www.sota2.com/research/paper/695de86653fdd5c3617439f3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77001e209131cea7066a.md>)

Recorded evaluation configuration: {"# Shots":"5-shot","# Activated Params":"37B","# Total Params":"671B"}. This record does not assert independent reproduction.

### MiMo-V2-Flash Base

Accuracy: 41.1. Result ID: 6996a26a5698665ec4ebf132.

[Reporting paper](<https://www.sota2.com/research/paper/695de86653fdd5c3617439f3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695de86653fdd5c3617439f3.md>)

[Implementation](<https://github.com/XiaomiMiMo/MiMo-V2-Flash>)

Recorded evaluation configuration: {"# Shots":"5-shot","# Activated Params":"15B","# Total Params":"309B"}. This record does not assert independent reproduction.

### Jet-RL

Accuracy: 35.2. Result ID: 69a188ea35dfcabb00e7b00c.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B-Base","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### BF16

Accuracy: 35.1. Result ID: 69a188ea35dfcabb00e7b00a.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B-Base","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### Initial

Accuracy: 31.8. Result ID: 69a188ea35dfcabb00e7b009.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B-Base","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### BF16-Train-FP8-Rollout

Accuracy: 30.3. Result ID: 69a188ea35dfcabb00e7b00b.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B-Base","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### BF16

Accuracy: 28.6. Result ID: 69a188ea35dfcabb00e7b007.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-7B","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### Jet-RL

Accuracy: 28.5. Result ID: 69a188ea35dfcabb00e7b008.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-7B","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### Initial

Accuracy: 25.5. Result ID: 69a188ea35dfcabb00e7b006.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-7B","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### Jet-RL

Accuracy: 19.9. Result ID: 69a188ea35dfcabb00e7b005.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Llama3.1-8B","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### BF16

Accuracy: 15.9. Result ID: 69a188ea35dfcabb00e7b003.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Llama3.1-8B","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### BF16-Train-FP8-Rollout

Accuracy: 14.7. Result ID: 69a188ea35dfcabb00e7b004.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Llama3.1-8B","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

### Initial

Accuracy: 12.4. Result ID: 69a188ea35dfcabb00e7b002.

[Reporting paper](<https://www.sota2.com/research/paper/69705dede25ad2fa9d1ed833.md>)

Recorded evaluation configuration: {"Model":"Llama3.1-8B","Rollout Length":"8k","Training Dataset (RL)":"GSM8k + MATH"}. This record does not assert independent reproduction.

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