# Multi-hop Question Answering on NQ (Accuracy) Source: [Multi-hop Question Answering on NQ (Accuracy)]() Collection index: [llms.txt]() Data updated: 2026-06-04T08:18:35.162Z Evaluation of model performance on the Natural Questions (NQ) dataset for multi-hop question answering, measured using Accuracy. 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 | Multi-hop Question Answering | | Dataset (including any stated split) | NQ | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 6 of 6 | | 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 - [Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Qwen2.5-7B-Instruct + MemTrain + MEM1 | 35.18 | Not recorded | 6a21065ec10484bac23efebf | | Qwen3-4B-Instruct-2507 + MemTrain + MEM1 | 32.24 | Not recorded | 6a21065ec10484bac23efebf | | Qwen2.5-7B-Instruct + MEM1 | 24.21 | Not recorded | 6a21065ec10484bac23efebf | | Qwen3-4B-Instruct-2507 + MEM1 | 21.97 | Not recorded | 6a21065ec10484bac23efebf | | Qwen3-4B-Instruct-2507 | 19.92 | Not recorded | 6a21065ec10484bac23efebf | | Qwen2.5-7B-Instruct | 12.22 | Not recorded | 6a21065ec10484bac23efebf | ### Qwen2.5-7B-Instruct + MemTrain + MEM1 Accuracy: 35.18. Result ID: 6a213331fbc373592c1ca447. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B-Instruct","Memory Tool":"MEM1","Training":"MemTrain"}. This record does not assert independent reproduction. ### Qwen3-4B-Instruct-2507 + MemTrain + MEM1 Accuracy: 32.24. Result ID: 6a213331fbc373592c1ca444. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Memory Tool":"MEM1","Training":"MemTrain"}. This record does not assert independent reproduction. ### Qwen2.5-7B-Instruct + MEM1 Accuracy: 24.21. Result ID: 6a213331fbc373592c1ca446. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B-Instruct","Memory Tool":"MEM1"}. This record does not assert independent reproduction. ### Qwen3-4B-Instruct-2507 + MEM1 Accuracy: 21.97. Result ID: 6a213331fbc373592c1ca443. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507","Memory Tool":"MEM1"}. This record does not assert independent reproduction. ### Qwen3-4B-Instruct-2507 Accuracy: 19.92. Result ID: 6a213331fbc373592c1ca442. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B-Instruct-2507"}. This record does not assert independent reproduction. ### Qwen2.5-7B-Instruct Accuracy: 12.22. Result ID: 6a213331fbc373592c1ca445. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B-Instruct"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.