KAT-V1: Kwai-AutoThink Technical Report

2025-07-11

AI Summary

To address the pervasive “overthinking” problem—redundant chain-of-thought invocation on simple inputs—in complex reasoning tasks, this paper introduces KAT-V1, the first open-source 40B large language model supporting dynamic reasoning-mode switching. Its core innovation is the AutoThink training paradigm: an intent-aware prompting mechanism estimates task complexity; majority-voting signals guide mode selection between reasoning and direct-response; and a dual-modality data construction strategy, cold-start initialization, and Step-SRPO reinforcement learning jointly enable fine-grained control and efficient transfer across modes. Experiments demonstrate that KAT-V1 matches or surpasses DeepSeek-R1-0528 and Qwen3-235B-A22B on multiple reasoning benchmarks while reducing reasoning token consumption by ~30%. The model has been deployed in Kwai’s programming assistant, Kwaipilot.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

We present Kwaipilot-AutoThink (KAT), an open-source 40B large language model developed to address the overthinking problem in reasoning-intensive tasks, where an automatic thinking training paradigm is proposed to dynamically switch between reasoning and non-reasoning modes based on task complexity. Specifically, first, we construct the dual-regime dataset based on a novel tagging pipeline and a multi-agent synthesis strategy, and then we apply Multi-Token Prediction (MTP)-enhanced knowledge distillation, enabling efficient and fine-grained reasoning transfer with minimal pretraining cost. Besides, we implement a cold-start initialization strategy that introduces mode-selection priors using majority-vote signals and intent-aware prompting. Finally, we propose Step-SRPO, a reinforcement learning algorithm that incorporates intermediate supervision into the GRPO framework, offering structured guidance over both reasoning-mode selection and response accuracy. Extensive experiments across multiple benchmarks demonstrate that KAT consistently matches or even outperforms current state-of-the-art models, including DeepSeek-R1-0528 and Qwen3-235B-A22B, across a wide range of reasoning-intensive tasks while reducing token usage by up to approximately 30%. Beyond academic evaluation, KAT has been successfully deployed in Kwaipilot (i.e., Kuaishou's internal coding assistant), and improves real-world development workflows with high accuracy, efficiency, and controllable reasoning behaviors. Moreover, we are actively training a 200B Mixture-of-Experts (MoE) with 40B activation parameters, where the early-stage results already demonstrate promising improvements in performance and efficiency, further showing the scalability of the AutoThink paradigm.

Problem

• Addresses overthinking in reasoning tasks with dynamic mode switching
• Enhances reasoning transfer via multi-token prediction knowledge distillation
• Improves efficiency and accuracy with structured RL-guided reasoning selection

Innovation

• Automatic thinking training paradigm
• Multi-Token Prediction knowledge distillation
• Step-SRPO reinforcement learning algorithm