ResearchBenchmarksMulti-task Language Understanding on MMLU-Pro (Avg@8)Follow73.76Avg@8TTRL68.393669.786871.1872.5732Jul 2, 2026Evaluation ResultsMethodMethodLinksAvg@8TTRLEval=crossEval=cross2026.0773.76N-OPSDEval=crossEval=cross2026.0773.74LMSIEval=crossEval=cross2026.0773.46IntuitorEval=crossEval=cross2026.0773.3Qwen3-4B base2026.0772.16TTRLEval=inEval=in2026.0772.04N-OPSDEval=inEval=in2026.0772.04IntuitorEval=inEval=in2026.0771.91LMSIEval=inEval=in2026.0768.6