# Long-context Question Answering on LongBench v2 (full) Source: [Long-context Question Answering on LongBench v2 (full)]() Collection index: [llms.txt]() Data updated: 2026-07-02T09:44:01.300Z Evaluation of Long-context Question Answering performance on the full configuration of the LongBench v2 dataset, reporting accuracy across overall, easy, hard, short, medium, and long context segments. 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 Question Answering | | Dataset (including any stated split) | LongBench v2 (full) | | Metric | Overall Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 10 of 10 | | 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 - [Overall Accuracy](): Higher is better; unit: PERCENT. - [Easy Accuracy](): Higher is better; unit: PERCENT. - [Hard Accuracy](): Higher is better; unit: PERCENT. - [Short Accuracy](): Higher is better; unit: PERCENT. - [Medium Accuracy](): Higher is better; unit: PERCENT. - [Long Accuracy](): Higher is better; unit: PERCENT. - [Latency (s)](): Lower is better; unit: TIME. - [Single-Document QA Accuracy](): Higher is better; unit: PERCENT. - [Multi-Document QA Accuracy](): Higher is better; unit: PERCENT. - [Long In-context Learning Accuracy](): Higher is better; unit: PERCENT. - [Long-dialogue History Understanding Accuracy](): Higher is better; unit: PERCENT. - [Code Repository Understanding Accuracy](): Higher is better; unit: PERCENT. - [Long Structured Data Understanding Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Overall Accuracy | Result date | Report paper | | --- | --- | --- | --- | | Full context | 56.7 | Not recorded | 6a1e640edbd7150506f85683 | | Full context (100k) | 51.1 | Not recorded | 6a1e640edbd7150506f85683 | | LongAttnComp (Stage 2, nosubq) | 49.7 | Not recorded | 6a1e640edbd7150506f85683 | | LongAttnComp (Stage 2, subq) | 48.9 | Not recorded | 6a1e640edbd7150506f85683 | | RLM | 46.7 | Not recorded | 6a45f096dc71372d7795a3bf | | MPLM | 46.5 | Not recorded | 6a45f096dc71372d7795a3bf | | Speculative Prefill | 46.3 | Not recorded | 6a1e640edbd7150506f85683 | | LongAttnComp (Stage 1) | 41.7 | Not recorded | 6a1e640edbd7150506f85683 | | MPLM | 37.8 | Not recorded | 6a45f096dc71372d7795a3bf | | RLM | 29.7 | Not recorded | 6a45f096dc71372d7795a3bf | ### Full context Overall Accuracy: 56.7. Result ID: 6a1eb608fbc373592c19b406. [Reporting paper]() Recorded evaluation configuration: {"Context length constraint":"Untruncated"}. This record does not assert independent reproduction. ### Full context (100k) Overall Accuracy: 51.1. Result ID: 6a1eb608fbc373592c19b407. [Reporting paper]() Recorded evaluation configuration: {"Context length constraint":"100k input tokens"}. This record does not assert independent reproduction. ### LongAttnComp (Stage 2, nosubq) Overall Accuracy: 49.7. Result ID: 6a1eb608fbc373592c19b40b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training stage":"Stage 2","Query-construction variant":"nosubq"}. This record does not assert independent reproduction. ### LongAttnComp (Stage 2, subq) Overall Accuracy: 48.9. Result ID: 6a1eb608fbc373592c19b40a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training stage":"Stage 2","Query-construction variant":"subq"}. This record does not assert independent reproduction. ### RLM Overall Accuracy: 46.7. Result ID: 6a4632dd60b88bb339864cb0. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen3.6-35B-A3B","Fine-tuning":"None"}. This record does not assert independent reproduction. ### MPLM Overall Accuracy: 46.5. Result ID: 6a4632dd60b88bb339864caf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3.6-35B-A3B","Fine-tuning":"None"}. This record does not assert independent reproduction. ### Speculative Prefill Overall Accuracy: 46.3. Result ID: 6a1eb608fbc373592c19b408. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Prefilling strategy":"Speculative"}. This record does not assert independent reproduction. ### LongAttnComp (Stage 1) Overall Accuracy: 41.7. Result ID: 6a1eb608fbc373592c19b409. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Training stage":"Stage 1"}. This record does not assert independent reproduction. ### MPLM Overall Accuracy: 37.8. Result ID: 6a4632dd60b88bb339864cad. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-30B-A3B","Fine-tuning":"None"}. This record does not assert independent reproduction. ### RLM Overall Accuracy: 29.7. Result ID: 6a4632dd60b88bb339864cae. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen3-30B-A3B","Fine-tuning":"None"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.