Mistral's Voxtral Mini 4B is a 4-billion parameter speech model that runs entirely in your browser — no cloud, no API keys, no cost.
Meta's Llama 4 introduces three open-weight AI models using Mixture-of-Experts architecture. Here's how Scout, Maverick, and Behemoth compare to Claude Opus 4.6, GPT-5.2, and DeepSeek V3.2.
DeepSeek V3.2 Speciale matches GPT-5.2 on math benchmarks while costing 10x less. Full review of China's most impressive open-source AI model.
AI-generated code floods GitHub but contributes nothing back. Maintainers are burning out. Is vibe coding killing the open-source ecosystem?
