Large Language Model News & Updates

Mistral has launched Devstral 2, a 123-billion parameter dense transformer optimized for agentic software development, along with a smaller variant, Devstral Small 2, and a command-line interface, Vibe CLI. Devstral 2 achieves 72.2% on SWE-bench Verified and is available for free for a limited time via Mistral's API and Hugging Face.

Z.ai has released an open-source vision-language model series, GLM-4.6V, for multimodal reasoning and frontend automation. The model achieves state-of-the-art results across various benchmarks, including general VQA, chart understanding, and STEM reasoning.

Booking.com doubled accuracy in key tasks using a disciplined, modular AI agent strategy, combining small models for inference, larger models for reasoning, and domain-tuned evaluations.

Nvidia and the University of Hong Kong release Orchestrator, an 8-billion-parameter model that coordinates tools and LLMs for complex problem-solving. This model outperforms larger models at a lower cost and improves efficiency.

Gemini 3 outperformed Copilot in most AI tasks, showcasing its strength in creative and knowledge-intensive tasks, while Copilot excelled in coding tasks.

Claude Code now supports fine-tuning of open-source large language models (LLMs) using Hugging Face skills, enabling users to fine-tune models with ease and speed up the development process.

We review the top 5 dictation software available for free and paid use, including Apple Dictation, Windows Voice Access, and more.