Moonshot AI

Latest AI news, models and releases from Moonshot AI. ['GLM 5.2', 'GLM-5.2', 'Hermes', 'K2', 'K2.6', 'K2.7', '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', 'kimi-k2-0711-preview', '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', 'Kimi K3.5', 'Kimi K4', 'Kimi K5', 'Kimi Linear', 'Kivine', 'Krea 2', 'Krea 2 Turbo', 'Mamba', 'Mamba-2', 'Mamba-3', 'MiniCPM-RobotManip', 'MiniCPM-RobotTrack']

Official site ↗

Open Source 🇷🇺

Twisterminigen: native Krea 2 Turbo studio on Swift and MLX for Mac

Developer created Twisterminigen, a native SwiftUI app for local Krea 2 Turbo inference on Mac mini M4 (32 GB). The app runs entirely without Python inside, uses MLX, and generates 1024×1024 images in about 7 minutes. A critical bug in the text encoder (wrong RoPE positions for 5 tokens) caused 99.32% pixel change after fix, despite near-perfect cosine similarity tests.

Moonshot AI

OpenAI

Alibaba/Qwen

Habr — хаб ИИ29.07 · 15:02

Models 🇨🇳

28 trillion tokens is not the limit: Chinese AI model Kimi K4 aims higher, GPUs are the bottleneck

Moonshot AI is developing a new model, Kimi K4, expected to surpass its predecessor Kimi K3 which was trained on 28.8 trillion tokens. The company's CEO indicated that the model's scale is limited by the availability of GPUs, highlighting hardware constraints as a bottleneck.

Moonshot AI

DeepSeek (GNews)29.07 · 14:02

Show more ↓