# Long-context reasoning on RULER 32K context length

Source: [Long-context reasoning on RULER 32K context length](<https://www.sota2.com/research/sota/long-context-reasoning-on-ruler-32k-context-length>)

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

Data updated: 2026-05-20T09:14:13.093Z

Evaluation of model performance on the RULER dataset specifically configured with a 32,000 token context length, focusing on long-context reasoning capabilities, measured by NIAH, AGG, QA, VT, and AVG scores.

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 | Long-context reasoning |
| Dataset (including any stated split) | RULER 32K context length |
| Metric | NIAH |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 3 of 3 |
| 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

- [NIAH](<https://www.sota2.com/research/sota/long-context-reasoning-on-ruler-32k-context-length.md?metric=NIAH>): Higher is better; unit: PERCENT.

- [AGG](<https://www.sota2.com/research/sota/long-context-reasoning-on-ruler-32k-context-length.md?metric=AGG>): Higher is better; unit: PERCENT.

- [QA](<https://www.sota2.com/research/sota/long-context-reasoning-on-ruler-32k-context-length.md?metric=QA>): Higher is better; unit: PERCENT.

- [VT](<https://www.sota2.com/research/sota/long-context-reasoning-on-ruler-32k-context-length.md?metric=VT>): Higher is better; unit: PERCENT.

- [AVG Score](<https://www.sota2.com/research/sota/long-context-reasoning-on-ruler-32k-context-length.md?metric=AVG%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | NIAH | Result date | Report paper |
| --- | --- | --- | --- |
| UltraLLaDA | 92.78 | Not recorded | 6a0d4054f984739ed6737b45 |
| UltraLLaDA+BA-Att | 90.97 | Not recorded | 6a0d4054f984739ed6737b45 |
| UltraLLaDA+XAtt | 74.94 | Not recorded | 6a0d4054f984739ed6737b45 |

### UltraLLaDA

NIAH: 92.78. Result ID: 6a0d7a04fbc373592c05b890.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4054f984739ed6737b45.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c0a1e209131cea8c08b.md>)

Recorded evaluation configuration: {"Base Model":"UltraLLaDA","Attention Method":"Full Attention"}. This record does not assert independent reproduction.

### UltraLLaDA+BA-Att

NIAH: 90.97. Result ID: 6a0d7a04fbc373592c05b892.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4054f984739ed6737b45.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d4054f984739ed6737b45.md>)

Recorded evaluation configuration: {"Base Model":"UltraLLaDA","Attention Method":"BA-Att"}. This record does not assert independent reproduction.

### UltraLLaDA+XAtt

NIAH: 74.94. Result ID: 6a0d7a04fbc373592c05b891.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4054f984739ed6737b45.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77811e209131cea73a62.md>)

Recorded evaluation configuration: {"Base Model":"UltraLLaDA","Attention Method":"XAtt"}. This record does not assert independent reproduction.

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