AI Newsletter

Mistral has released Mistral Large 4 (codename 'Le Chonk') as an API preview: 1.05 trillion parameters, 49 billion active, 1M token context, trained on 4,000 Nvidia Blackwell GPUs. Open weights to follow on October 27 – Mistral is explicitly positioning itself as the Western answer to Chinese open-weight models. Meanwhile, Mistral CEO Arthur Mensch raised 3 billion euros in September, underscoring ambitions for sovereign AI infrastructure.

OpenAI has published 722 mathematical papers created by an internal frontier model – from approximately 4,000 worked problems. The publication was coordinated with an independent advisory group for mathematics and AI; many results still lack computer-verified proofs. The move is seen as evidence that AI systems can independently generate scientific insights – a quality leap with significant relevance for research and competition.

JetBrains – longtime market leader in developer tools – reports its first net loss in company history. Analyst Gergely Orosz attributes this directly to the changed demand structure caused by AI coding agents. The case is viewed as a warning signal for established software vendors who have missed the transition to agentic workflows.

The BCG Applied AI Index 2026 shows: Nearly 50% of companies are achieving measurable value creation through AI, AI spending has nearly doubled to 3.3% of revenue, and agentic AI is increasingly being deployed productively. This shifts the debate from hype to concrete scaling – companies still in the exploration phase risk falling structurally behind.

IMF Managing Director describes three simultaneous shocks to the global economy: rapid AI diffusion, high energy prices, and record government debt. According to the IMF, the interactions of these three factors determine which economies benefit from AI transformation and which are destabilized. For companies, this means increased planning uncertainty for infrastructure and AI investments.