ResearchTasksLong-Context InferenceLong-Context InferenceFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedLongBench >128K tokens v2 (long)SnapKV (Best)35.2Accuracy8Feb 26, 2026LongBench 32K~128K tokens v2 (medium)H2O (Best)28.8Accuracy8Feb 26, 2026LongBench <32K tokens v2 (short)H2O (Best)41.7Accuracy8Feb 26, 2026LongBench v2 (hard)H2O (Best)30.2Accuracy8Feb 26, 2026Long-context memory-constrained setting 16 docs x 8K tokens, 10 GB KV poolLazy-Attn7.2Time To First Token (TTFT) (s)2Jun 4, 2026
Long-context memory-constrained setting 16 docs x 8K tokens, 10 GB KV poolLazy-Attn7.2Time To First Token (TTFT) (s)2Jun 4, 2026