# Long-context language understanding on LongBench (7-Metric Subset: NQ, HQ, QM, MN, GR, TQ, Avg)

Source: [Long-context language understanding on LongBench (7-Metric Subset: NQ, HQ, QM, MN, GR, TQ, Avg)](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg>)

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

Data updated: 2026-04-23T10:03:16.450Z

Evaluation of long-context language understanding on the LongBench suite, reporting performance across a specific subset of 7 key metrics including NQ, HQ, QM, MN, GR, TQ, and the overall Average.

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 |
| Metric | NQ Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| 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

- [NQ Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg.md?metric=NQ%20Score>): Higher is better; unit: PERCENT.

- [HQ Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg.md?metric=HQ%20Score>): Higher is better; unit: PERCENT.

- [QM Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg.md?metric=QM%20Score>): Higher is better; unit: PERCENT.

- [MN Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg.md?metric=MN%20Score>): Higher is better; unit: PERCENT.

- [GR Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg.md?metric=GR%20Score>): Higher is better; unit: PERCENT.

- [TQ Score](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg.md?metric=TQ%20Score>): Higher is better; unit: PERCENT.

- [Average Score (LongBench Subset)](<https://www.sota2.com/research/sota/long-context-language-understanding-on-longbench-7-metric-subset-nq-hq-qm-mn-gr-tq-avg.md?metric=Average%20Score%20(LongBench%20Subset)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | NQ Score | Result date | Report paper |
| --- | --- | --- | --- |
| DASH-KV | 29.52 | Not recorded | 69e85bed7d0172140ec811de |
| Full Attn | 28.2 | Not recorded | 69e85bed7d0172140ec811de |
| Full Attn | 28.11 | Not recorded | 69e85bed7d0172140ec811de |
| DASH-KV | 27.61 | Not recorded | 69e85bed7d0172140ec811de |
| Full Attn | 25.13 | Not recorded | 69e85bed7d0172140ec811de |
| DASH-KV | 24.65 | Not recorded | 69e85bed7d0172140ec811de |
| H2O | 24.4 | Not recorded | 69e85bed7d0172140ec811de |
| SnapKV | 24.1 | Not recorded | 69e85bed7d0172140ec811de |
| SnapKV | 23.86 | Not recorded | 69e85bed7d0172140ec811de |
| H2O | 23.1 | Not recorded | 69e85bed7d0172140ec811de |
| H2O | 22.88 | Not recorded | 69e85bed7d0172140ec811de |
| SnapKV | 21.3 | Not recorded | 69e85bed7d0172140ec811de |
| StreamLLM | 20.47 | Not recorded | 69e85bed7d0172140ec811de |
| StreamLLM | 18.3 | Not recorded | 69e85bed7d0172140ec811de |
| StreamLLM | 13.6 | Not recorded | 69e85bed7d0172140ec811de |

### DASH-KV

NQ Score: 29.52. Result ID: 69e9ee5147f34ce924d1a12d.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B-Instruct","Type":"Retrieval"}. This record does not assert independent reproduction.

### Full Attn

NQ Score: 28.2. Result ID: 69e9ee5147f34ce924d1a129.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B-Instruct","Type":"Dense"}. This record does not assert independent reproduction.

### Full Attn

NQ Score: 28.11. Result ID: 69e9ee5147f34ce924d1a124.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Type":"Dense"}. This record does not assert independent reproduction.

### DASH-KV

NQ Score: 27.61. Result ID: 69e9ee5147f34ce924d1a128.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Type":"Retrieval"}. This record does not assert independent reproduction.

### Full Attn

NQ Score: 25.13. Result ID: 69e9ee5147f34ce924d1a11f.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2-7B-Instruct","Type":"Dense"}. This record does not assert independent reproduction.

### DASH-KV

NQ Score: 24.65. Result ID: 69e9ee5147f34ce924d1a123.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2-7B-Instruct","Type":"Retrieval"}. This record does not assert independent reproduction.

### H2O

NQ Score: 24.4. Result ID: 69e9ee5147f34ce924d1a12b.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718c1e209131cea51ead.md>)

[Implementation](<https://github.com/FMInference/H2O>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B-Instruct","Type":"Eviction"}. This record does not assert independent reproduction.

### SnapKV

NQ Score: 24.1. Result ID: 69e9ee5147f34ce924d1a12c.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B-Instruct","Type":"Retrieval"}. This record does not assert independent reproduction.

### SnapKV

NQ Score: 23.86. Result ID: 69e9ee5147f34ce924d1a122.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Qwen2-7B-Instruct","Type":"Retrieval"}. This record does not assert independent reproduction.

### H2O

NQ Score: 23.1. Result ID: 69e9ee5147f34ce924d1a126.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718c1e209131cea51ead.md>)

[Implementation](<https://github.com/FMInference/H2O>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Type":"Eviction"}. This record does not assert independent reproduction.

### H2O

NQ Score: 22.88. Result ID: 69e9ee5147f34ce924d1a121.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa718c1e209131cea51ead.md>)

[Implementation](<https://github.com/FMInference/H2O>)

Recorded evaluation configuration: {"Backbone":"Qwen2-7B-Instruct","Type":"Eviction"}. This record does not assert independent reproduction.

### SnapKV

NQ Score: 21.3. Result ID: 69e9ee5147f34ce924d1a127.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73b21e209131cea5df89.md>)

[Implementation](<https://github.com/FasterDecoding/SnapKV>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Type":"Retrieval"}. This record does not assert independent reproduction.

### StreamLLM

NQ Score: 20.47. Result ID: 69e9ee5147f34ce924d1a120.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72d71e209131cea58d79.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2-7B-Instruct","Type":"Eviction"}. This record does not assert independent reproduction.

### StreamLLM

NQ Score: 18.3. Result ID: 69e9ee5147f34ce924d1a12a.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72d71e209131cea58d79.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B-Instruct","Type":"Eviction"}. This record does not assert independent reproduction.

### StreamLLM

NQ Score: 13.6. Result ID: 69e9ee5147f34ce924d1a125.

[Reporting paper](<https://www.sota2.com/research/paper/69e85bed7d0172140ec811de.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72d71e209131cea58d79.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Type":"Eviction"}. This record does not assert independent reproduction.

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