Moonshot AI

Latest AI news, models and releases from Moonshot AI. ['GLM-5.2', 'Hermes', 'K3', 'K3-mini', 'K3-MoE', 'Kimi', 'Kimi 2.6', 'Kimi-2.6', 'Kimi 3', 'Kimi 3T', 'Kimi Hosted Agent', 'Kimi k1.5', 'Kimi K1.5', 'Kimi K2', 'Kimi K2.5', 'Kimi-K2.5-1T-A32B', 'Kimi K2.6', 'Kimi K2.7', 'Kimi K2.7 Code', 'Kimi K2 Thinking', 'Kimi K3', 'Kimi K3.5', 'Kimi K5', 'Kimi Linear', 'Kivine', 'Mamba', 'Mamba-2', 'Mamba-3', 'MiniCPM-RobotManip', 'MiniCPM-RobotTrack', 'Minimax', 'Ornith‑1.0‑35B', 'Qwen3-0.6B', 'Qwen3-30B-A3B', 'Qwen3-4B-Instruct-2507', 'qwen3.5-9b', 'Strands Agents', 'модель с триллионом параметров']

Official site ↗

Models 🇷🇺

How Kimi K3 Works: 2.8 Trillion Parameters, Linear Attention, and Million-Token Agents

Moonshot AI has released the weights of its Kimi K3 model, a multimodal model based on a Mixture of Experts architecture with 2.8 trillion parameters, of which 104 billion are activated per token. In certain programming and tool-use benchmarks, the model outperforms competitors, but on average it lags behind some of them.

Moonshot AI

Anthropic

OpenAI

Habr — хаб ИИ27.07 · 23:03

Research 🇺🇸

Managing Reasoning Levels in Large Language Models

Sebastian Raschka explains how reasoning modes of varying complexity (low, medium, high) work in large language models. He discusses reinforcement learning methods with verifiable rewards, as well as mechanisms for switching between modes, including approaches used in models like DeepSeek-R1, Qwen3, and GPT-5.6.

OpenAI

DeepSeek

Moonshot AI

Sebastian Raschka27.07 · 20:04

Show more ↓