Elite AI researchers are defecting between top labs like OpenAI, Anthropic, and Meta at record rates, fueled by massive compensation packages, ideological clashes, and a race for computing power and influence — even as leaders warn the talent war may undermine mission-driven work before AI automates these roles entirely.
A new open-source Python framework called 'smevals' launches, offering developers a structured, systematic way to evaluate AI language model performance through organized tests, automated grading, and built-in reporting tools.
A new open-source Rust library called pdf-inspector now classifies and extracts PDF text in under 200ms without OCR or ML models, supporting Python, Node.js, and WebAssembly, and outperforming all tested local engines across 200 PDFs — completing the full corpus in just 0.47 seconds.
OpenAI's internal AI model, Astra, achieves stunning mathematical breakthroughs by solving or advancing ten long-standing open problems across fields including geometry, group theory, and quantum complexity, with landmark results including disproving Connes's rigidity conjecture and proving new bounds for Ramsey numbers.
Anthropic's claude-opus-5-max claims the top spot on the WebDev AI Leaderboard with a record Elo score of 1705 across 510,000 votes, while open-source models from DeepSeek and Z.ai push into the top 10, signaling a tightening race between proprietary and open-source AI in web development.
A consortium of AI companies including Loka, AWS, and Arcee AI releases Trinity Mini, an open 26-billion-parameter biomedical reasoning model trained with reinforcement learning that achieves 81.2% drug evaluation accuracy, offering organizations an auditable, deployable solution for specialized scientific AI within controlled infrastructure.
Anthropic's Claude Mythos AI has dealt a fatal blow to the post-quantum signature scheme HAWK by producing a successful key recovery attack, effectively ending its standardization prospects, while also achieving a modest improvement on reduced-round AES attacks that poses no real-world threat to current encryption.
Google DeepMind has quietly disbanded its Nobel Prize-winning AlphaFold team less than a year after its historic 2024 Nobel Prize, with key researchers including Nobel co-winner John Jumper defecting to rival Anthropic, which just launched Claude Science to target the biology and drug-discovery space AlphaFold helped pioneer.
Lilian Weng departs Thinking Machines due to health concerns and returns to OpenAI, where she will lead a top-level research team focused on accelerating internal research, including recursive self-improvement.
OpenAI's GPT-2 small weights are outperforming custom-trained models on instruction-following tasks despite worse test loss scores, sparking investigation into whether superior data quality, overtraining, or dropout are behind the surprising gap.