The quiet spell is officially over. A new tech giant just dropped its own open models, while others released AI that can reason, see, and even watch videos. Here’s what’s new for your local setup.

Just when we thought the market was settling, the game changed again. This isn’t just an incremental update with fine-tunes; this is a look at the future of local AI, with new players and entirely new capabilities hitting the scene.

Here’s the latest intel on our almanac update:

• 🔹 Before: 71 models
• 🔹 New Drops: 4
• 🔸 New Total: 75 models

Let’s break down the new powerhouses you can run on your own hardware right now.

🔥 The Corporate Pro: Granite-4 (3B) by IBM

Out of nowhere, tech giant IBM just crashed the open-source party with its “Granite” model family. The 3B version is a compact, highly efficient model trained specifically for enterprise tasks. This isn’t for writing poems; it’s a sharp, focused tool for business logic, document analysis, and corporate workflows. A serious new player has entered the ring.

• VRAM: ~4GB
• Size: 2.1GB
• Ollama Name: granite4:3b

🧠 The Thinker: DeepSeek V3.2-Exp (Reasoning)

This is a game-changer. DeepSeek’s new experimental model features a “thinking mode.” Before generating a response, it creates an internal chain-of-thought or a reasoning plan. This dramatically improves its performance on complex, multi-step problems, making it one of the best local models for deep analysis and logical challenges. It doesn’t just answer; it thinks.

• VRAM: ~16GB
• Size: 9.0GB
• Ollama Name: deepseek-v3.2-exp

🎬 The Everything Model: Qwen3-Next/Omni

Hold onto your hats. This is the true multimodal beast we’ve been waiting for. Qwen3-Omni is a “Swiss Army knife” that natively understands not just text and images, but also audio and video. You can feed it a short video clip and start asking questions about what’s happening. For local AI, this is a massive leap forward.

• VRAM: ~20GB
• Size: 15.0GB
• Ollama Name: qwen3-omni

💻 The Next-Gen Coder: Code-Llama 4 (15B)

It was only a matter of time. The community has already dropped the first major coding fine-tune on the powerful Llama 4 Scout architecture. This specialized version is built to be a next-generation coding assistant, showing incredible promise in autocompletion, debugging, and understanding complex codebases.

• VRAM: ~10GB
• Size: 8.5GB
• Ollama Name: codellama4:15b

Ready to Test the Future?

The era of simple text-based chatbots is ending. The next wave of local AI is here, with models that can reason, analyze, and interact with the world in entirely new ways.

You can run ALL of these groundbreaking models with our LLM VibeCheck app.

➡️ Download LLM VibeCheck, grab the latest almanac, and get your hands on the next generation of AI!

 

Follow the Vibe