# Long-context language understanding on LongBench 128k

Source: [Long-context language understanding on LongBench 128k](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-128k>)

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

Data updated: 2026-05-11T09:24:08.299Z

Evaluation of long-context language understanding capabilities using the LongBench dataset configured specifically for a 128,000 token context window, reporting results across key QA and summarization sub-tasks.

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) | LongBench 128k |
| Metric | HotpotQA |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 2 of 2 |
| 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

- [HotpotQA](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-128k.md?metric=HotpotQA>): Higher is better; unit: PERCENT.

- [NarrativeQA](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-128k.md?metric=NarrativeQA>): Higher is better; unit: PERCENT.

- [TriviaQA](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-128k.md?metric=TriviaQA>): Higher is better; unit: PERCENT.

- [TREC](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-128k.md?metric=TREC>): Higher is better; unit: PERCENT.

- [MultiNews](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-128k.md?metric=MultiNews>): Higher is better; unit: PERCENT.

- [Qasper](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-128k.md?metric=Qasper>): Higher is better; unit: PERCENT.

## Recorded results

| Method | HotpotQA | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen2.5 | 57.96 | Not recorded | 68fa7c421e209131cea8d35b |
| SpikingBrain-76B | 48.42 | Not recorded | 68fa7c421e209131cea8d35b |

### Qwen2.5

HotpotQA: 57.96. Result ID: 6a01a01d47f34ce924e3c96a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c421e209131cea8d35b.md>)

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

Recorded evaluation configuration: {"Params":"7B","Complexity Type":"Quadratic"}. This record does not assert independent reproduction.

### SpikingBrain-76B

HotpotQA: 48.42. Result ID: 6a01a01d47f34ce924e3c969.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7c421e209131cea8d35b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c421e209131cea8d35b.md>)

Recorded evaluation configuration: {"Params":"12B/76B","Complexity Type":"Hybrid"}. This record does not assert independent reproduction.

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