# Mathematical Reasoning on ARC-c and GPQA-Diamond

Source: [Mathematical Reasoning on ARC-c and GPQA-Diamond](<https://www.sota2.com/research/sota/mathematical-reasoning-on-arc-c-and-gpqa-diamond>)

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

Data updated: 2026-04-13T08:19:49.119Z

Evaluation of model performance on the Out-of-Distribution reasoning components ARC-c and GPQA-Diamond, focusing on Mathematical Reasoning.

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) | Out-of-Distribution Reasoning Suite ARC-c, GPQA-Diamond |
| Metric | ARC-c (pass@1) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 14 of 14 |
| 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

- [ARC-c (pass@1)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-arc-c-and-gpqa-diamond.md?metric=ARC-c%20(pass%401)>): Higher is better; unit: PERCENT.

- [GPQA-D (pass@1)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-arc-c-and-gpqa-diamond.md?metric=GPQA-D%20(pass%401)>): Higher is better; unit: PERCENT.

- [Average Score](<https://www.sota2.com/research/sota/mathematical-reasoning-on-arc-c-and-gpqa-diamond.md?metric=Average%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | ARC-c (pass@1) | Result date | Report paper |
| --- | --- | --- | --- |
| RL | 82.3 | Not recorded | 69dc785a7d0172140ec788bd |
| DYPO | 81.8 | Not recorded | 69dc785a7d0172140ec788bd |
| SRFT | 81.6 | Not recorded | 69dc785a7d0172140ec788bd |
| CHORD | 81.1 | Not recorded | 69dc785a7d0172140ec788bd |
| LUFFY | 80.5 | Not recorded | 69dc785a7d0172140ec788bd |
| SuperRL | 77.8 | Not recorded | 69dc785a7d0172140ec788bd |
| SFT | 75.2 | Not recorded | 69dc785a7d0172140ec788bd |
| ReLIFT | 74.9 | Not recorded | 69dc785a7d0172140ec788bd |
| PRIME-Zero | 73.3 | Not recorded | 69dc785a7d0172140ec788bd |
| SFT → RL | 72.4 | Not recorded | 69dc785a7d0172140ec788bd |
| Oat-Zero | 70.1 | Not recorded | 69dc785a7d0172140ec788bd |
| OpenReasoner-Zero | 66.2 | Not recorded | 69dc785a7d0172140ec788bd |
| SimpleRL-Zero | 30.2 | Not recorded | 69dc785a7d0172140ec788bd |
| Qwen2.5-Math-7B | 18.2 | Not recorded | 69dc785a7d0172140ec788bd |

### RL

ARC-c (pass@1): 82.3. Result ID: 69dca5cc47f34ce924c383aa.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6bf71e209131cea314c0.md>)

[Implementation](<https://github.com/openai/following-instructions-human-feedback>)

Recorded evaluation configuration: {"Training Category":"Reinforcement Learning","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### DYPO

ARC-c (pass@1): 81.8. Result ID: 69dca5cc47f34ce924c383b5.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

Recorded evaluation configuration: {"Training Category":"SFT and RL","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### SRFT

ARC-c (pass@1): 81.6. Result ID: 69dca5cc47f34ce924c383b3.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79ab1e209131cea7ecd7.md>)

Recorded evaluation configuration: {"Training Category":"SFT and RL","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### CHORD

ARC-c (pass@1): 81.1. Result ID: 69dca5cc47f34ce924c383b4.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bfc1e209131cea8ba85.md>)

Recorded evaluation configuration: {"Training Category":"SFT and RL","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### LUFFY

ARC-c (pass@1): 80.5. Result ID: 69dca5cc47f34ce924c383b1.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79961e209131cea7e682.md>)

Recorded evaluation configuration: {"Training Category":"SFT and RL","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### SuperRL

ARC-c (pass@1): 77.8. Result ID: 69dca5cc47f34ce924c383b0.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a851e209131cea83abd.md>)

Recorded evaluation configuration: {"Training Category":"SFT and RL","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### SFT

ARC-c (pass@1): 75.2. Result ID: 69dca5cc47f34ce924c383a9.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c131e209131cea31f71.md>)

Recorded evaluation configuration: {"Training Category":"Supervised Fine-Tuning","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### ReLIFT

ARC-c (pass@1): 74.9. Result ID: 69dca5cc47f34ce924c383b2.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7be61e209131cea8b28c.md>)

Recorded evaluation configuration: {"Training Category":"SFT and RL","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### PRIME-Zero

ARC-c (pass@1): 73.3. Result ID: 69dca5cc47f34ce924c383ad.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b9f1e209131cea8965d.md>)

Recorded evaluation configuration: {"Training Category":"Reinforcement Learning","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### SFT → RL

ARC-c (pass@1): 72.4. Result ID: 69dca5cc47f34ce924c383af.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa701f1e209131cea492fe.md>)

[Implementation](<https://github.com/facebookresearch/llama>)

Recorded evaluation configuration: {"Training Category":"SFT and RL","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### Oat-Zero

ARC-c (pass@1): 70.1. Result ID: 69dca5cc47f34ce924c383ae.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7be51e209131cea8b201.md>)

Recorded evaluation configuration: {"Training Category":"Reinforcement Learning","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### OpenReasoner-Zero

ARC-c (pass@1): 66.2. Result ID: 69dca5cc47f34ce924c383ac.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79d91e209131cea7fba6.md>)

Recorded evaluation configuration: {"Training Category":"Reinforcement Learning","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### SimpleRL-Zero

ARC-c (pass@1): 30.2. Result ID: 69dca5cc47f34ce924c383ab.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a721e209131cea8345b.md>)

Recorded evaluation configuration: {"Training Category":"Reinforcement Learning","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

### Qwen2.5-Math-7B

ARC-c (pass@1): 18.2. Result ID: 69dca5cc47f34ce924c383a8.

[Reporting paper](<https://www.sota2.com/research/paper/69dc785a7d0172140ec788bd.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa767e1e209131cea6d6a7.md>)

Recorded evaluation configuration: {"Training Category":"Base Model","Backbone":"Qwen2.5-Math-7B","Inference Temperature":"0.6","Option Shuffling":"true"}. This record does not assert independent reproduction.

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