Jul 28, 2025 · We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip …

Join the discussion on this paper pagearXiv explained breakdown of this paper 👉 https://arxivexplained.com/papers/kimi-k2-open-agentic-intelligence

View recent discussion. Abstract: We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We …

Jul 11, 2025 · Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters. Trained with the Muon optimizer, …

Aug 4, 2025 · In this video, we review Kimi K2 from MoonshotAI. https://arxiv.org/pdf/2507.20534 Kimi K2 is trained with agentic use in mind.

DOI: 10.48550/ARXIV.2507.20534 access: open type: Informal or Other Publication metadata version: 2025-09-04

Abstract We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, …

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which …

Kimi K2 is an open-source trillion-parameter MoE model advancing agentic intelligence via innovative design, scalable training, and robust RL performance.

Benchmarking Kimi K2 The table below details the performance of Kimi-K2-Instruct, showing that it matches—or outperforms—the latest open-source and proprietary models across a diverse …

Jul 28, 2025 · We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip …

This paper discusses a specific research topic in the field of science or technology, providing insights and findings relevant to the subject.