Open Weights

Mistral Large 4: 1.05 trillion parameters, 52B active per token. We measured our 4× DGX Spark cluster — 23.0 tok/s, 42.2 aggregate, €0.553 per 1M output tokens.

Ternary Bonsai 2 compresses Qwen3.8-27B to 5.95 GB — 98.2% of the quality. What that means for 16 GB and 8 GB setups, with our own cluster measurement.

GLM-5.3-Flash, DeepSeek V4.1 Flash and Qwen3.8-Flash-Next in overview — and the honest hardware matrix: what runs on a laptop, Mac Studio, DGX Spark and multi-GPU node?

OpenAI has terminated its contract with Cursor effective November 2026, triggering a major industry shift. Following SpaceX's acquisition of Cursor, this case proves that relying on a single AI model provider is a strategic liability. Here is why builders need multi-provider architectures and strict ToS reviews.

NVIDIA has confirmed the acquisition of Hugging Face for $12.93 billion. Discover what this means for the future of open-weights infrastructure and the practical steps tech teams should take to secure their model supply chains.

Moonshot published Kimi K3's weights on 27 July 2026 under its own licence, not MIT. What the three commercial clauses mean before you run a benchmark.

Clem Delangue asked OpenAI to release the rogue agent traces and commit $100M. The part that matters for defenders: hosted safety filters blocked the forensics.