ResearchBenchmarksQuestion Answering on LongBench Code Repo QA Long v2Question Answering on LongBench Code Repo QA Long v2Follow72.41AccuracyBabelTele43.716451.165758.61566.0643Jun 18, 2026Evaluation resultsMethodMethodLinksAccuracyBabelTeleModel=GLM-5.1Model=GLM-5.12026.0672.41OriginalModel=GLM-5.1Model=GLM-5.12026.0662.07BabelTeleModel=Qwen3.6 MaxModel=Qwen3.6 Max2026.0662.07OriginalModel=Qwen3.6 MaxModel=Qwen3.6 Max2026.0655.17BabelTeleModel=Kimi2.5Model=Kimi2.52026.0648.28OriginalModel=Kimi2.5Model=Kimi2.52026.0644.82