# Long-context language understanding on RULER 32K

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

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

Data updated: 2026-05-26T12:36:45.061Z

Evaluation of model performance on the RULER dataset configured for a 32,000 token context length, reporting aggregated scores across Easy, Hard, and Full categories.

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 language understanding |
| Dataset (including any stated split) | RULER 32K |
| Metric | Easy Average Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

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

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

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

## Recorded results

| Method | Easy Average Score | Result date | Report paper |
| --- | --- | --- | --- |
| HashAttention | 85.8 | Not recorded | 68fa7bf01e209131cea8b723 |
| oracle-top-k | 85.31 | Not recorded | 68fa7bf01e209131cea8b723 |
| full attention | 85.3 | Not recorded | 68fa7bf01e209131cea8b723 |
| vAttention(oracle-top-k) | 85.17 | Not recorded | 68fa7bf01e209131cea8b723 |
| vAttention(HashAttention) | 84.57 | Not recorded | 68fa7bf01e209131cea8b723 |

### HashAttention

Easy Average Score: 85.8. Result ID: 6a1593d3fbc373592c0e5ab5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bf01e209131cea8b723.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa74a51e209131cea62e6e.md>)

Recorded evaluation configuration: {"Model":"Meta-llama/Llama-3.1-8B-Instruct","Context Window":"32K","Sparsity":"10%"}. This record does not assert independent reproduction.

### oracle-top-k

Easy Average Score: 85.31. Result ID: 6a1593d3fbc373592c0e5ab3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bf01e209131cea8b723.md>)

Recorded evaluation configuration: {"Model":"Meta-llama/Llama-3.1-8B-Instruct","Context Window":"32K","Sparsity":"10%"}. This record does not assert independent reproduction.

### full attention

Easy Average Score: 85.3. Result ID: 6a1593d3fbc373592c0e5ab1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bf01e209131cea8b723.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa703d1e209131cea49db1.md>)

[Implementation](<https://github.com/tensorflow/tensor2tensor>)

Recorded evaluation configuration: {"Model":"Meta-llama/Llama-3.1-8B-Instruct","Context Window":"32K"}. This record does not assert independent reproduction.

### vAttention(oracle-top-k)

Easy Average Score: 85.17. Result ID: 6a1593d3fbc373592c0e5ab2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bf01e209131cea8b723.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bf01e209131cea8b723.md>)

Recorded evaluation configuration: {"Model":"Meta-llama/Llama-3.1-8B-Instruct","Context Window":"32K","Sparsity":"10%","Approximation":"oracle-top-k"}. This record does not assert independent reproduction.

### vAttention(HashAttention)

Easy Average Score: 84.57. Result ID: 6a1593d3fbc373592c0e5ab4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bf01e209131cea8b723.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bf01e209131cea8b723.md>)

Recorded evaluation configuration: {"Model":"Meta-llama/Llama-3.1-8B-Instruct","Context Window":"32K","Sparsity":"10%","Approximation":"HashAttention"}. This record does not assert independent reproduction.

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