# Question Answering on NQ (EM and Latency)

Source: [Question Answering on NQ (EM and Latency)](<https://www.sota2.com/research/sota/question-answering-on-nq-em-and-latency>)

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

Data updated: 2026-05-08T10:42:46.563Z

Evaluation of model performance and latency on the Natural Questions (NQ) dataset for Question Answering, reporting Exact Match (EM) and response time.

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 | Question Answering |
| Dataset (including any stated split) | NQ |
| Metric | EM (%) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 27 of 27 |
| 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

- [EM (%)](<https://www.sota2.com/research/sota/question-answering-on-nq-em-and-latency.md?metric=EM%20(%25)>): Higher is better; unit: PERCENT.

- [Latency (ms)](<https://www.sota2.com/research/sota/question-answering-on-nq-em-and-latency.md?metric=Latency%20(ms)>): Lower is better; unit: TIME.

## Recorded results

| Method | EM (%) | Result date | Report paper |
| --- | --- | --- | --- |
| LatentRAG△ | 49.86 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG♢ | 49.03 | Not recorded | 69fd6e205497b3b670c894ce |
| AutoRefine△ | 48.2 | Not recorded | 69fd6e205497b3b670c894ce |
| Search-R1♢ | 48.03 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG△ | 47.59 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG♢ | 47.37 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG♢ | 46.29 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG♢ | 46.15 | Not recorded | 69fd6e205497b3b670c894ce |
| AutoRefine△ | 45.98 | Not recorded | 69fd6e205497b3b670c894ce |
| Search-R1♢ | 45.79 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG△ | 45.73 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG△ | 45.68 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG♢ | 45.54 | Not recorded | 69fd6e205497b3b670c894ce |
| LatentRAG△ | 45.32 | Not recorded | 69fd6e205497b3b670c894ce |
| Search-R1♢ | 44.4 | Not recorded | 69fd6e205497b3b670c894ce |
| AutoRefine△ | 44.35 | Not recorded | 69fd6e205497b3b670c894ce |
| Search-R1♢ | 43.93 | Not recorded | 69fd6e205497b3b670c894ce |
| AutoRefine△ | 43.82 | Not recorded | 69fd6e205497b3b670c894ce |
| AutoRefine△ | 43.8 | Not recorded | 69fd6e205497b3b670c894ce |
| Search-R1♢ | 43.35 | Not recorded | 69fd6e205497b3b670c894ce |
| R1-Searcher | 36.34 | Not recorded | 69fd6e205497b3b670c894ce |
| ZeroSearch | 34.07 | Not recorded | 69fd6e205497b3b670c894ce |
| DeepRAG | 33.16 | Not recorded | 69fd6e205497b3b670c894ce |
| Iter-RetGen | 31.33 | Not recorded | 69fd6e205497b3b670c894ce |
| Naive RAG | 25.84 | Not recorded | 69fd6e205497b3b670c894ce |
| Search-o1 | 22.33 | Not recorded | 69fd6e205497b3b670c894ce |
| Direct Infer | 14.76 | Not recorded | 69fd6e205497b3b670c894ce |

### LatentRAG△

EM (%): 49.86. Result ID: 69fdbe1d47f34ce924e16e0d.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"e5-base-v2"}. This record does not assert independent reproduction.

### LatentRAG♢

EM (%): 49.03. Result ID: 69fdbe1d47f34ce924e16e0b.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"e5-base-v2"}. This record does not assert independent reproduction.

### AutoRefine△

EM (%): 48.2. Result ID: 69fdbe1d47f34ce924e16e0c.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88288.md>)

Recorded evaluation configuration: {"Retrieval Model":"e5-base-v2"}. This record does not assert independent reproduction.

### Search-R1♢

EM (%): 48.03. Result ID: 69fdbe1d47f34ce924e16e0a.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a651e209131cea82e7b.md>)

Recorded evaluation configuration: {"Retrieval Model":"e5-base-v2"}. This record does not assert independent reproduction.

### LatentRAG△

EM (%): 47.59. Result ID: 69fdbe1d47f34ce924e16e11.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"jina-embeddings-v5-text-nano"}. This record does not assert independent reproduction.

### LatentRAG♢

EM (%): 47.37. Result ID: 69fdbe1d47f34ce924e16e0f.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"jina-embeddings-v5-text-nano"}. This record does not assert independent reproduction.

### LatentRAG♢

EM (%): 46.29. Result ID: 69fdbe1d47f34ce924e16e07.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### LatentRAG♢

EM (%): 46.15. Result ID: 69fdbe1d47f34ce924e16e13.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"harrier-oss-v1-270m"}. This record does not assert independent reproduction.

### AutoRefine△

EM (%): 45.98. Result ID: 69fdbe1d47f34ce924e16e10.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88288.md>)

Recorded evaluation configuration: {"Retrieval Model":"jina-embeddings-v5-text-nano"}. This record does not assert independent reproduction.

### Search-R1♢

EM (%): 45.79. Result ID: 69fdbe1d47f34ce924e16e0e.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a651e209131cea82e7b.md>)

Recorded evaluation configuration: {"Retrieval Model":"jina-embeddings-v5-text-nano"}. This record does not assert independent reproduction.

### LatentRAG△

EM (%): 45.73. Result ID: 69fdbe1d47f34ce924e16e09.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### LatentRAG△

EM (%): 45.68. Result ID: 69fdbe1d47f34ce924e16e15.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"harrier-oss-v1-270m"}. This record does not assert independent reproduction.

### LatentRAG♢

EM (%): 45.54. Result ID: 69fdbe1d47f34ce924e16e17.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"F2LLM-v2-330M"}. This record does not assert independent reproduction.

### LatentRAG△

EM (%): 45.32. Result ID: 69fdbe1d47f34ce924e16e19.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"F2LLM-v2-330M"}. This record does not assert independent reproduction.

### Search-R1♢

EM (%): 44.4. Result ID: 69fdbe1d47f34ce924e16e12.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a651e209131cea82e7b.md>)

Recorded evaluation configuration: {"Retrieval Model":"harrier-oss-v1-270m"}. This record does not assert independent reproduction.

### AutoRefine△

EM (%): 44.35. Result ID: 69fdbe1d47f34ce924e16e08.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88288.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### Search-R1♢

EM (%): 43.93. Result ID: 69fdbe1d47f34ce924e16e06.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a651e209131cea82e7b.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### AutoRefine△

EM (%): 43.82. Result ID: 69fdbe1d47f34ce924e16e18.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88288.md>)

Recorded evaluation configuration: {"Retrieval Model":"F2LLM-v2-330M"}. This record does not assert independent reproduction.

### AutoRefine△

EM (%): 43.8. Result ID: 69fdbe1d47f34ce924e16e14.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7b641e209131cea88288.md>)

Recorded evaluation configuration: {"Retrieval Model":"harrier-oss-v1-270m"}. This record does not assert independent reproduction.

### Search-R1♢

EM (%): 43.35. Result ID: 69fdbe1d47f34ce924e16e16.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a651e209131cea82e7b.md>)

Recorded evaluation configuration: {"Retrieval Model":"F2LLM-v2-330M"}. This record does not assert independent reproduction.

### R1-Searcher

EM (%): 36.34. Result ID: 69fdbe1d47f34ce924e16e03.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77771e209131cea7356d.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### ZeroSearch

EM (%): 34.07. Result ID: 69fdbe1d47f34ce924e16e04.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa789d1e209131cea79568.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### DeepRAG

EM (%): 33.16. Result ID: 69fdbe1d47f34ce924e16e05.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79411e209131cea7ce63.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### Iter-RetGen

EM (%): 31.33. Result ID: 69fdbe1d47f34ce924e16e01.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70f21e209131cea4e48a.md>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### Naive RAG

EM (%): 25.84. Result ID: 69fdbe1d47f34ce924e16e00.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69b91e209131cea23c26.md>)

[Implementation](<https://github.com/huggingface/transformers/blob/master/examples/rag/>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### Search-o1

EM (%): 22.33. Result ID: 69fdbe1d47f34ce924e16e02.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76911e209131cea6dca9.md>)

[Implementation](<https://github.com/sunnynexus/Search-o1>)

Recorded evaluation configuration: {"Retrieval Model":"Qwen3-Embedding-0.6B"}. This record does not assert independent reproduction.

### Direct Infer

EM (%): 14.76. Result ID: 69fdbe1d47f34ce924e16dff.

[Reporting paper](<https://www.sota2.com/research/paper/69fd6e205497b3b670c894ce.md>)

Recorded evaluation configuration: {"Retrieval Model":"N/A"}. This record does not assert independent reproduction.

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