# Addition on 12-digit addition 3 terms

Source: [Addition on 12-digit addition 3 terms](<https://www.sota2.com/research/sota/addition-on-12-digit-addition-3-terms>)

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

Data updated: 2026-06-04T09:44:52.922Z

Evaluation of model performance on the Addition task using the 12-digit addition 3 terms dataset, measured by Token Accuracy, Question Accuracy, Modification Rate, TP Correction, and FP Preservation.

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 | Addition |
| Dataset (including any stated split) | 12-digit addition 3 terms |
| Metric | Token Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| 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

- [Token Accuracy](<https://www.sota2.com/research/sota/addition-on-12-digit-addition-3-terms.md?metric=Token%20Accuracy>): Higher is better; unit: PERCENT.

- [Question Accuracy](<https://www.sota2.com/research/sota/addition-on-12-digit-addition-3-terms.md?metric=Question%20Accuracy>): Higher is better; unit: PERCENT.

- [Modification Rate](<https://www.sota2.com/research/sota/addition-on-12-digit-addition-3-terms.md?metric=Modification%20Rate>): Lower is better; unit: PERCENT.

- [TP Correction](<https://www.sota2.com/research/sota/addition-on-12-digit-addition-3-terms.md?metric=TP%20Correction>): Higher is better; unit: PERCENT.

- [FP Preservation](<https://www.sota2.com/research/sota/addition-on-12-digit-addition-3-terms.md?metric=FP%20Preservation>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Token Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| REPLACEMENT | 95.47 | Not recorded | 6a2106a8c10484bac23f033c |
| STEERING | 94.7 | Not recorded | 6a2106a8c10484bac23f033c |
| DUAL-STREAM PROBE | 94.62 | Not recorded | 6a2106a8c10484bac23f033c |
| DUAL-STREAM PROBE | 93.78 | Not recorded | 6a2106a8c10484bac23f033c |
| ORIGINAL | 93.29 | Not recorded | 6a2106a8c10484bac23f033c |
| DUAL-STREAM PROBE | 91.87 | Not recorded | 6a2106a8c10484bac23f033c |
| RE-PROMPTING | 88.06 | Not recorded | 6a2106a8c10484bac23f033c |

### REPLACEMENT

Token Accuracy: 95.47. Result ID: 6a214908fbc373592c1d4e3a.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B"}. This record does not assert independent reproduction.

### STEERING

Token Accuracy: 94.7. Result ID: 6a214908fbc373592c1d4e39.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76e51e209131cea6fbea.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B"}. This record does not assert independent reproduction.

### DUAL-STREAM PROBE

Token Accuracy: 94.62. Result ID: 6a214908fbc373592c1d4e3e.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Implementation](<https://github.com/RL-MIND/Shape-of-Addition>)

Recorded evaluation configuration: {"Model":"Qwen3-8B","Delta (δ)":"0.1"}. This record does not assert independent reproduction.

### DUAL-STREAM PROBE

Token Accuracy: 93.78. Result ID: 6a214908fbc373592c1d4e3d.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Implementation](<https://github.com/RL-MIND/Shape-of-Addition>)

Recorded evaluation configuration: {"Model":"Qwen3-8B","Delta (δ)":"0.05"}. This record does not assert independent reproduction.

### ORIGINAL

Token Accuracy: 93.29. Result ID: 6a214908fbc373592c1d4e38.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B"}. This record does not assert independent reproduction.

### DUAL-STREAM PROBE

Token Accuracy: 91.87. Result ID: 6a214908fbc373592c1d4e3c.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Implementation](<https://github.com/RL-MIND/Shape-of-Addition>)

Recorded evaluation configuration: {"Model":"Qwen3-8B","Delta (δ)":"0"}. This record does not assert independent reproduction.

### RE-PROMPTING

Token Accuracy: 88.06. Result ID: 6a214908fbc373592c1d4e3b.

[Reporting paper](<https://www.sota2.com/research/paper/6a2106a8c10484bac23f033c.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a081e209131cea80ab6.md>)

Recorded evaluation configuration: {"Model":"Qwen3-8B"}. This record does not assert independent reproduction.

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