Five brand-new AI models released in the last 30 days, all free, all open. You can run them locally right now. Let me break each one down fast. Number one, Gemma 4 from Google DeepMind. Apache 2.0 license, fully commercial, no restrictions. Four model sizes. The E2B variant runs in just 4 GB of RAM. That is a phone. The 31B dense model beats models 20 times its size on benchmarks. Supports text, images, audio, and video. Best for edge devices, multimodal agents, and low VRAM rigs. Number two, GLM 5.1 from Zep AI. MIT license, open weights. 744 billion parameters, trained entirely on Huawei Ascend chips, no Nvidia required. Here is the killer stat. SWE Bench Pro, 58.4. That is a new state-of-the-art, beating both GPT 5.4 and Claude Opus 4.6. Best

for agentic coding, long horizon tasks, and self-hosted enterprise. Number three, GLM 5V Turbo, the multimodal sibling. Feed it a screenshot or a Figma mockup, and it returns working code. It drives full GUI automation, real browsers, real terminals, full perception-to-execution loops. Best for design-to-code, visual coding agents, and GUI automation. Number four, Qwen 3.6 Plus from Alibaba. 1 million token context. That is 2,000 pages in a single prompt. Free on OpenRouter right now. SWE Bench verified at 78.8. Built on chain of thought that stays focused across hundreds of agent steps. Best for entire codebase analysis, long document reasoning, and complex agents. Number five, my personal favorite, Bonsai 8B by Prism ML out of Caltech. Apache 2.0, true one-bit architecture. The entire model is 1.15 GB. It runs on

an iPhone 17 Pro at 44 tokens per second. On an RTX 4090, it hits 368. Competitive benchmarks with full precision 8B models at 14 times less memory. Best for ultra-low VRAM, mobile, and CPU-only machines. Quick [snorts] recap. Gemma 4 for edge and multimodal. GLM 5.1 for frontier-level coding agents. GLM 5V Turbo for visual coding. Qwen 3.6 Plus for massive context. Bonsai 8B for running AI literally anywhere, even your phone. All free, all this month. Links to every model are in the description. Drop a comment telling me which one you're running first, and subscribe. This space is moving faster than ever, and I drop these breakdowns every week.