Publications

We are open sourcing our latest model, Kimi K2.6, featuring state-of-the-art coding, long-horizon execution, and agent swarm …

An agentic RL method that evolves the skill library through a fitness-driven lifecycle, enabling skills and the model to co-evolve throughout training.

We propose PRISM-Δ, a differential subspace steering method for prompt highlighting that matches or exceeds the best existing method on 19 of 20 configurations with relative gains up to +10.6%, while halving the fluency cost.

Probing hidden states during reasoning reveals that LLMs already know the answer before finishing thinking. We detect overthinking via ‘jumps’ and intervene during inference to improve reasoning.

Abstract: We present PromptCD, a test-time method for controlling LLM behavior without additional training. The approach creates paired positive and negative guiding prompts for a target behavior and contrasts model responses at the token-probability level for LLMs and through visual attention patterns for VLMs.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. The model focuses on …

A training-free framework that softly anchors fused predictions toward the target modality’s own judgment when overshooting, mitigating cross-modal hallucination in multimodal LLMs.

Abstract: Long contexts challenge transformers as attention scores dilute across thousands of tokens and critical information is often lost in the middle. We reframe test-time adaptation as a budget-constrained memory consolidation problem and propose Gdwm (Gated Differentiable Working Memory), which introduces a write controller that estimates Contextual Utility, an information-theoretic measure of long-range contextual dependence, to allocate gradient steps efficiently.

Abstract: Reinforcement learning (RL) has shown great promise in enhancing LLM reasoning, but current approaches mainly focus on single domains with verifiable rewards. We propose RGR-GRPO, a rubric-driven RL framework for multi-domain reasoning that uses rubrics to provide fine-grained reward signals and offline guidance.

The advancement of large language models (LLMs) has catalyzed a paradigm shift from code generation assistance to autonomous coding …

Vision-Language Models (VLMs) have demonstrated remarkable success across diverse visual tasks, yet their performance degrades in …

Abstract: Prompt-based adversarial attacks have become an effective means to assess the robustness of large language models (LLMs). However, existing approaches often treat prompts as monolithic text, overlooking their structural heterogeneity-different prompt components contribute unequally to adversarial robustness.

We present the first comprehensive analysis of how the order of graph descriptions impacts LLM performance, evaluating four graph description orders across six graph problems using six mainstream LLMs.

We introduce Kimi K2, a Mixture-of-Experts large language model with 32 billion activated parameters and 1 trillion total parameters. …

The performance of Large Language Models (LLMs) is fundamentally determined by the contextual information provided during inference. …

Abstract: Multimodal Retrieval-Augmented Generation (RAG) systems have become essential in knowledge-intensive and open-domain tasks. As retrieval complexity increases, ensuring the robustness of these systems is critical. However, current RAG models are highly sensitive to the order in which evidence is presented, often resulting in unstable performance and biased reasoning, particularly as the number of retrieved items or modality diversity grows.

Abstract: Large language models (LLMs) have achieved remarkable progress, demonstrating unprecedented capabilities across various natural language processing tasks. However, the high costs associated with such exceptional performance limit the widespread adoption of LLMs, highlighting the need for prompt compression.

Large Language Models (LLMs) have made remarkable breakthroughs in reasoning, yet continue to struggle with hallucinations, logical …

We present the first comprehensive analysis of the impacts of CoT prompting on Reasoning LLMs, finding that one-shot CoT consistently enhances performance and reduces excessive reflections by approximately 90%.

Abstract: The current research on text-guided 3D synthesis predominantly utilizes complex diffusion models, posing significant challenges in tasks like terrain generation. This study ventures into the direct synthesis of text-to-3D terrain in a zero-shot fashion, circumventing the need for diffusion models.

In this work, we propose a Dynamically Adaptive Density Control Strategy based on the degree of reconstruction of the background of the scene, which adaptive the spatial sample point generation strategy dynamically according to the training results and prevents the generation of redundant data in the model.

Adversarial attack for time-series classification model is widely explored and many attack methods are proposed. But there is not a …