Anthropic said Sunday it will put 5% of its shares into a new Anthropic Public Benefit Fund, subject to stockholder approval, and the board can add 2 more points for every 50% rise in the company’s value above a baseline it hasn’t disclosed. The fund is an LLC rather than a charity, so it can give to nonprofits or pay people directly, and Anthropic names AI-driven job losses as one target. With an IPO reportedly planned for November, this is both mission work and pre-listing positioning. Elon Musk replied by pledging one SpaceX share to every kid in America. Anthropic
MiniMax launched M3.1 Flash on October 19 and confirmed it was the anonymous “Space Bunny” model that sat at the top of OpenRouter usage for 10 days. It takes text, images and video with a 1M-token context, costs $0.15 per million input tokens and $0.60 per million output, and open weights are due this week. Developers picked it before they knew whose it was, which says more than the benchmark table. TokenPost
A Stanford, UIUC and Together AI paper on “Hybrid Reasoning Agents,” which do much of their thinking in the model’s hidden states and switch to text only for tool calls and key steps, drew one of the biggest AI threads on Hacker News. On Qwen3 models up to 14B, the authors report about 10% better scores with 35% to 55% fewer tokens and roughly 2x faster inference, and say the hidden steps can still be read back as text. If that holds up, agents get cheaper to run, but how much of their reasoning stays visible becomes a real question. arXiv
Leave a comment