# Path-Lock Expert: Separating Reasoning Mode in Hybrid Thinking via Architecture-Level Separation Source: [Path-Lock Expert: Separating Reasoning Mode in Hybrid Thinking via Architecture-Level Separation]() Collection index: [llms.txt]() Data updated: 2026-08-12T05:00:13.408Z Authors: Shouren Wang, Wang Yang, Chuang Ma, Debargha Ganguly, Vikash Singh, Chaoda Song, Xinpeng Li, Xianxuan Long, Vipin Chaudhary, Xiaotian Han Published: 2026-04-29T21:07:47.000Z ## Abstract Hybrid-thinking language models expose explicit /think and /no\_think modes, but current designs do not separate them cleanly. Even in /no\_think mode, models often emit long and self-reflective responses, causing reasoning leakage. Existing work reduces this issue through better data curation and multi-stage training, yet leakage remains because both modes are still encoded in the same feed-forward parameters. We propose Path-Lock Expert (PLE), an architecture-level solution that replaces the single MLP in each decoder layer with two semantically locked experts, one for /think and one for /no\_think, while keeping attention, embeddings, normalization, and the language-model head shared. A deterministic control-token router selects exactly one expert path for the entire sequence, so inference preserves the dense model's per-token computation pattern and each expert receives mode-pure updates during supervised fine-tuning. Across math and science reasoning benchmarks, PLE maintains strong /think performance while producing a substantially stronger mode separation, a /no\_think mode with higher accuracy and far less reasoning leakage. On Qwen3-4B, for example, compared to the SFT-only baseline on AIME24, PLE generates 17x fewer reflective tokens (6.01 vs. 0.35 per answer) and 2x shorter outputs (8665 vs. 4101 tokens), and improves /no\_think accuracy from 35.33% to 44.67%, while maintaining /think-mode performance (61.33% vs. 60.00%). These results suggest that controllable hybrid thinking is fundamentally an architectural problem, and separating mode-specific feed-forward pathways is a simple and effective solution. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Mathematical Reasoning on MATH500 (Acc., Len., #Relf./Ans.)](): Mathematical Reasoning; MATH500; Accuracy: 94.8. - [Question Answering on MMLU STEM (Acc, Len, Refl/Ans)](): Question Answering; MMLU STEM; Accuracy: 92.22. - [Mathematical Reasoning on AIME 24 (Accuracy, Length, #Relf./Ans.)](): Mathematical Reasoning; AIME 24; Accuracy: 60. - [Question Answering on GPQA Diamond (Accuracy, Length, Reflection Count)](): Question Answering; GPQA Diamond; Accuracy: 48. ## Original sources and implementation [Original paper]() [PDF]()