ResearchBenchmarksLong-context Question Answering on LongBench Long In-context Learning v2For AI assistantsMarkdownContent indexLong-context Question Answering on LongBench Long In-context Learning v2Follow49.4AccuracyMPLM25.58431.76737.9544.133Jul 1, 2026Evaluation resultsMethodMethodLinksAccuracyLatency (s)MPLMBackbone=Qwen3.6-35B-A3BBackbone=Qwen3.6-35B-A3B2026.0749.4118.1RLMBackbone=Qwen3.6-35B-A3BBackbone=Qwen3.6-35B-A3B2026.0738.3327.6MPLMBackbone=Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.0734.968.3RLMBackbone=Qwen3-30B-A3BBackbone=Qwen3-30B-A3B2026.0726.598.5