超越正確性：混合思考多模態大語言模型中回應行為的基準測試與對齊

摘要

混合思考多模態大型語言模型（MLLMs）允許單一模型在深思型思考與低延遲的非思考推論之間切換。雖然這些模式在推理預算上有所不同，但其產出的回應應滿足相同的使用者導向標準。僅靠正確性可能不足以刻畫這種回應品質；因此我們將任務準確度與回應模式失敗視為互補的評估結果。我們透過回應模式對齊來研究這個差距：即思考與非思考介面是否維持可接受的最終回應行為。我們提出 PatternEval，一個富含失敗案例的診斷基準，包含 2,415 個多模態提示，涵蓋視覺感知與接地、結構化影像理解，以及多模態知識推理。PatternEval 測試四種常見失敗：思維鏈洩漏、回應重複、邏輯矛盾，以及表演式推理。回應模式失敗在不同提供者的模型中普遍存在，其中非思考推論的失敗率明顯更高，因而在思考與非思考介面之間造成系統性的失對齊。基於此診斷，我們開發了 PatternRM，一個回應層級的獎勵模型，以及 PatternRL，其在強化學習期間引入特定模式的懲罰。在 Qwen3-VL-4B 與 Qwen3-VL-8B 上的實驗顯示，將特定模式懲罰納入強化學習可減輕跨模式失對齊，同時僅付出輕微的任務效能取捨。總結而言，PatternEval 與 PatternRL 提供了一個評估與訓練框架，用於在混合思考介面之間對齊使用者可見的回應模式。

English

Hybrid-thinking multimodal large language models (MLLMs) allow a single model to alternate between deliberative thinking and latency-efficient non-thinking inference. Although these modes differ in reasoning budget, their delivered responses should satisfy the same user-facing standard. Correctness alone may not characterize this response quality; we therefore evaluate task accuracy and response-pattern failures as complementary outcomes. We study this gap through response-pattern alignment: whether thinking and non-thinking interfaces preserve acceptable final-response behavior. We introduce PatternEval, a failure-enriched diagnostic benchmark comprising 2,415 multimodal prompts spanning visual perception and grounding, structured image understanding, and multimodal knowledge reasoning. PatternEval tests four recurrent failures: chain-of-thought leakage, response repetition, logical contradiction, and performative reasoning. Response-pattern failures are widespread across models from different providers, with non-thinking inference exhibiting substantially higher failure rates and thereby creating systematic misalignment between thinking and non-thinking interfaces. Motivated by this diagnosis, we develop PatternRM, a response-level reward model, and PatternRL, which introduces pattern-specific penalties during reinforcement learning. Experiments on Qwen3-VL-4B and Qwen3-VL-8B show that incorporating pattern-specific penalties into reinforcement learning can mitigate cross-mode misalignment while incurring a marginal task performance trade-off. Together, PatternEval and PatternRL provide an evaluation-and-training framework for aligning user-visible response patterns across hybrid-thinking interfaces.