ResearchBenchmarksOpen-ended Generation on Human Evaluation ScoresFollow4.4Quality ScoreSTA3.2563.5533.854.147Feb 4, 2026Evaluation ResultsMethodMethodLinksQuality ScoreAlignment ScoreSTABackbone=Gemma2-9B-itBackbone=Gemma2-9B-it2026.024.44.7AUSteerBackbone=Gemma2-9B-itBackbone=Gemma2-9B-it2026.024.34.7AUSteerBackbone=Qwen3-8BBackbone=Qwen3-8B2026.024.34.1SADIBackbone=Gemma2-9B-itBackbone=Gemma2-9B-it2026.024.24.5SADIBackbone=Qwen3-8BBackbone=Qwen3-8B2026.024.13.9AUSteerBackbone=LLaMA2-7B-ChatBackbone=LLaMA2-7B-Chat2026.023.43.8SADIBackbone=LLaMA2-7B-ChatBackbone=LLaMA2-7B-Chat2026.023.33.6