Hold up! The AI game just shifted gears with some massive drops this week. Meta officially unleashed the first wave of Llama 4 models, and the coding AI scene is getting seriously competitive.

Here’s the intel:
🔹 Before: 69 models
🔹 New Drops/Refreshed: 4
🔸 New Total: 71 models

Let’s dive into the heavy hitters:

🔥 The Flagship Drop: Llama 4 Maverick (80B) The new king has arrived! “Maverick” at a whopping 80 billion parameters is Meta’s flagship. This beast delivers top-tier performance in complex reasoning, mathematical challenges, and maintaining super-long conversations. This is your go-to for pushing the limits of local AI.

• VRAM: ~50GB
• Size: 45.0GB
• Ollama Name: llama4:80b

🔥 Agent Focus: Llama 4 Scout (15B) First hit the new breed! This isn’t about raw size. “Scout” is Meta’s new 15B model specifically designed for agentic workflows. It can plan tasks, use tools, and execute multi-step instructions autonomously. Your new personal AI assistant for automation.

• VRAM: ~10GB
• Size: 8.5GB
• Ollama Name: llama4:15b

💻 Code Master: Qwen 3 Coder (32B) Alibaba is stepping up! This 32B model is a direct competitor to Devstral. It’s custom-trained for code analysis, writing, and fixing bugs across multiple files. If you’re a developer, this model is about to make your life a lot easier.

• VRAM: ~22GB
• Size: 18.5GB
• Ollama Name: qwen3:32b-coder

✨ Logic & Wizardry: WizardLM 3 (Mistral 2 22B) The community’s back with a supercharged fine-tune. WizardLM is known for its ability to follow complex instructions and tackle tricky logical problems. Built on the powerful Mistral 2 22B base, this version is possibly the smartest model in the sub-30B class.

• VRAM: ~14GB
• Size: 12.5GB
• Ollama Name: wizardlm3

The full almanac is updated and ready to unleash this new power. You can use ALL these incredible models with our LLM VibeCheck app. Go download them and push the boundaries of what’s possible!

Follow the Vibe