ResearchBenchmarksLong-context Question Answering on LongBench I. Single-Document QA v2For AI assistantsMarkdownContent indexLong-context Question Answering on LongBench I. Single-Document QA v2Follow48AccuracyMPLM35.5238.764245.24Jul 1, 2026Evaluation resultsMethodMethodLinksAccuracyLatency (s)MPLMBackbone=Qwen3.6-35B-A3BBackbone=Qwen3.6-35B-A3B2026.0748107.9RLMBackbone=Qwen3.6-35B-A3BBackbone=Qwen3.6-35B-A3B2026.0747.4233MPLMBackbone=Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.0738.158.2RLMBackbone=Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.0736117.3