Token-Level Length Control: 3B Model Beats GPT 5.4 and Claude
token级,精准控制生成长度:3B模型击败GPT 5.4、Claude
UC Santa Barbara and Apple researchers introduced LenVM, which models remaining generation length as a token-level value function; Qwen2.5-3B with a 1.5B LenVM scored 62.6 on LIFEBench length control, above GPT-5.4 at 37.4 and Claude-Opus-4-6 at 35.5.
Why it matters: Modeling remaining generation length as a token-level value function lets a 3B model beat GPT-5.4 and Claude on length control.