Open-source AI models
6 picks we've reviewed.
Qwen-AgentWorld
Alibaba's Qwen team open-sources AgentWorld — a language world model that simulates seven agent environments (MCP, Search, Terminal, SWE, Web, OS, Android) and is trained to predict how each environment responds to an action. Released June 24, 2026 under Apache 2.0.
GLM-5.2
Zhipu's open-weight coding flagship: a 753B mixture-of-experts model with a 1M-token context and MIT weights, claiming to edge past GPT-5.5 on coding benchmarks.
VibeThinker-3B
A 3-billion-parameter open reasoning model that claims to match systems hundreds of times its size on math and code — and has the AI world arguing about whether the benchmarks are real.
Gemma 4
Google DeepMind's fourth-generation open-weight model family — five sizes from 2B to 31B, Apache 2.0 licensed, with the 12B Unified variant accepting text, image, audio, and video in a single encoder-free architecture.
Pyramid Flow
An open-source, MIT-licensed text- and image-to-video model that makes 768p, 24fps, 10-second clips — and runs on a single consumer GPU with offloading.
OpenHands
An open, self-hosted control center for autonomous coding agents — own the infrastructure, bring your own model, and run agents that edit files and ship code.