# Long-context retrieval on RULER 64K context Source: [Long-context retrieval on RULER 64K context]() Collection index: [llms.txt]() Data updated: 2026-04-28T10:21:50.936Z Evaluation of model performance on the RULER dataset specifically configured for a 64,000 token context length, focusing on Long-context retrieval accuracy and delta metrics. 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 retrieval | | Dataset (including any stated split) | RULER 64K context | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 19 of 19 | | 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. - [Delta](): Higher is better; unit: SCALAR. - [Correct Count](): Higher is better; unit: SCALAR. - [Accuracy Std Dev](): Lower is better; unit: SCALAR. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | WindowedManifoldKV | 84.3 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Windowed-4K | 84.29 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Windowed-8K | 83.92 | Not recorded | 698abbd9c97a0b3e3b5de23d | | WindowedManifoldKV | 83.9 | Not recorded | 698abbd9c97a0b3e3b5de23d | | WindowedManifoldKV | 82.4 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Windowed-16K | 82.4 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Hybrid (0.3) | 81.26 | Not recorded | 698abbd9c97a0b3e3b5de23d | | KeyDiff | 81.1 | Not recorded | 698abbd9c97a0b3e3b5de23d | | KeyDiff | 81.09 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Normalized | 81.06 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Hybrid (0.5) | 79.14 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Full Attention | 78.16 | Not recorded | 69f03f7b55a3c32af89fa561 | | KSA | 76.09 | Not recorded | 69f03f7b55a3c32af89fa561 | | Hybrid-KSA | 76.04 | Not recorded | 69f03f7b55a3c32af89fa561 | | Hybrid-SWA | 73.88 | Not recorded | 69f03f7b55a3c32af89fa561 | | Hybrid-Linear | 73.5 | Not recorded | 69f03f7b55a3c32af89fa561 | | Hybrid-SCA | 68.88 | Not recorded | 69f03f7b55a3c32af89fa561 | | ManifoldKV | 35.2 | Not recorded | 698abbd9c97a0b3e3b5de23d | | Global ManifoldKV | 35.2 | Not recorded | 698abbd9c97a0b3e3b5de23d | ### WindowedManifoldKV Accuracy: 84.3. Result ID: 6998daf95698665ec43f01d6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Type":"Local L2","Window Size":"4K"}. This record does not assert independent reproduction. ### Windowed-4K Accuracy: 84.29. Result ID: 6998daf95698665ec43f01db. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%","Window Size":"4K"}. This record does not assert independent reproduction. ### Windowed-8K Accuracy: 83.92. Result ID: 6998daf95698665ec43f01dc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%","Window Size":"8K"}. This record does not assert independent reproduction. ### WindowedManifoldKV Accuracy: 83.9. Result ID: 6998daf95698665ec43f01d7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Type":"Local L2","Window Size":"8K"}. This record does not assert independent reproduction. ### WindowedManifoldKV Accuracy: 82.4. Result ID: 6998daf95698665ec43f01d8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Type":"Local L2","Window Size":"16K"}. This record does not assert independent reproduction. ### Windowed-16K Accuracy: 82.4. Result ID: 6998daf95698665ec43f01dd. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%","Window Size":"16K"}. This record does not assert independent reproduction. ### Hybrid (0.3) Accuracy: 81.26. Result ID: 6998daf95698665ec43f01de. [Reporting paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%","Hybrid Alpha":0.3}. This record does not assert independent reproduction. ### KeyDiff Accuracy: 81.1. Result ID: 6998daf95698665ec43f01d9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Type":"Cosine"}. This record does not assert independent reproduction. ### KeyDiff Accuracy: 81.09. Result ID: 6998daf95698665ec43f01df. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%"}. This record does not assert independent reproduction. ### Normalized Accuracy: 81.06. Result ID: 6998daf95698665ec43f01e0. [Reporting paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%"}. This record does not assert independent reproduction. ### Hybrid (0.5) Accuracy: 79.14. Result ID: 6998daf95698665ec43f01e1. [Reporting paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%","Hybrid Alpha":0.5}. This record does not assert independent reproduction. ### Full Attention Accuracy: 78.16. Result ID: 69f08a2b47f34ce924d723bd. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Setting":"CPT setting"}. This record does not assert independent reproduction. ### KSA Accuracy: 76.09. Result ID: 69f08a2b47f34ce924d723c1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"CPT setting"}. This record does not assert independent reproduction. ### Hybrid-KSA Accuracy: 76.04. Result ID: 69f08a2b47f34ce924d723c2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"CPT setting"}. This record does not assert independent reproduction. ### Hybrid-SWA Accuracy: 73.88. Result ID: 69f08a2b47f34ce924d723be. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Setting":"CPT setting"}. This record does not assert independent reproduction. ### Hybrid-Linear Accuracy: 73.5. Result ID: 69f08a2b47f34ce924d723c0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"CPT setting"}. This record does not assert independent reproduction. ### Hybrid-SCA Accuracy: 68.88. Result ID: 69f08a2b47f34ce924d723bf. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Setting":"CPT setting"}. This record does not assert independent reproduction. ### ManifoldKV Accuracy: 35.2. Result ID: 6998daf95698665ec43f01da. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Type":"Global L2"}. This record does not assert independent reproduction. ### Global ManifoldKV Accuracy: 35.2. Result ID: 6998daf95698665ec43f01e2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Samples":6497,"Compression Ratio":"25%","Window Mode":"Global"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.