ResearchBenchmarksQuestion Answering on BABILong (8K to 1M Context Lengths)Question Answering on BABILong (8K to 1M Context Lengths)Follow63.54QA Accuracy (8K Context)MHM-LRU58.662459.928761.19562.4613Jul 1, 2026Evaluation resultsMethodMethodLinksQA Accuracy (8K Context)QA Accuracy (16K Context)QA Accuracy (32K Context)QA Accuracy (64K Context)QA Accuracy (128K Context)QA Accuracy (256K Context)QA Accuracy (512K Context)QA Accuracy (1M Context)MHM-LRUMethod Type=Multi-Head...Method Type=Multi-Head Recurrent Memory2026.0763.5460.9455.4751.8248.747.3946.3541.41ReMemMethod Type=Recurrent...Method Type=Recurrent Memory2026.0761.4655.2145.5742.1934.6432.2929.524.48MemAgentMethod Type=Recurrent...Method Type=Recurrent Memory2026.0758.8555.9953.3947.1444.6638.0232.5525.26