2026-10-07
By Vadym · Generated with AI, curated by me
Today splits into two camps: companies betting that more compute wins, and companies betting that users are tired of AI reaching into their data. Both bets showed up at once.
Mistral AI announced Mistral Large 4, internally nicknamed “Le Chonk,” a multimodal model with 1 trillion parameters aimed at both closed American models and Chinese open-weight rivals. VP of Science Pierre Stock said it was trained on roughly 4,000 Nvidia GPUs — “two to three times less” than Chinese competitors use. It’s available now only through a guardrailed endpoint; open weights follow in three weeks after safety testing.
Lambda is raising as much as $4 billion at a $14.5 billion pre-money valuation, led by Coatue and Blackstone, pushing its planned IPO to 2027. Its order backlog jumped from $15 billion in June to $50 billion by September — though much of that growth traces to a single $35 billion commitment from Anthropic signed in late August.
Claude for Startups gives eligible companies — founded within five years or funded within two — a year of Claude Team (up to five seats), $1,000 in API credits, Marketplace access, and office hours with Anthropic’s Applied AI team. Anthropic framed it directly: “the benefits of AI will reach most people through the companies that build on top of models, rather than through the models alone.”
Hark Pro, from a startup founded by Brett Adcock, uses a model trained specifically for computer use to handle tasks like expense approvals and DMV renewals, showing its browsing moves in a small window as it works. The pitch leans hard on privacy: the company says it’s “not here to like sell you ads and steal your data,” a direct jab at Meta and OpenAI. It launches today with a free tier and a paid tier for heavy use.
The San Francisco startup skips LLMs for a custom foundation model that tracks how consumer behavior shifts over time, not just a static snapshot of it. CEO Abhivyakti Ahuja says the goal is to “capture the longitudinal data on how people are changing, what triggers are changing them.” In one case, the model told a pet food brand its packaging wasn’t the problem — a perception of cheapness was. The two-year-old company is closing its first venture round and competes at TechCrunch Disrupt’s Startup Battlefield this month.
Lambda is raising billions to build more compute, and Anthropic is handing out credits to get more builders onto its models — the scale-up bet is as loud as ever. But LibreOffice refusing to ship AI by default, and Hark selling itself on what it won’t do with your data, are both aimed at the same anxiety: people don’t distrust AI’s capability, they distrust what happens to their information once they grant access. The companies that answer that question honestly, rather than just promising it, are the ones worth watching next.
— Boba
Curated by Vadym