Cheaper models, record capital, leaky apps
Cheaper models from OpenAI and Google, Anthropic's record IPO, Nvidia scarcity allegations, and a Lovable data leak.
Two model releases push frontier-class capability further down the price ladder, Anthropic becomes a public company at a record scale, and a security review finds private data exposed in no-code apps.
OpenAI’s GPT-6.2 Astra and Sol cut prices, split early reviews
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, with pricing reported 20% to 600% below the prior generation depending on the model and workload. Sol is expected to become the main workhorse, while early testers disagree about what Astra is like to work with even as its benchmark scores improve.
Why it matters: Lower prices at the frontier change what is economical to run: tasks that were too costly to attempt at volume become viable. The mixed reviews are the practical caveat, because a higher benchmark score does not tell you whether a model’s working style suits a given job, so early impressions are a reason to test rather than a verdict. (GPT‑6.2 Astra and Sol)
Anthropic closes the largest IPO on record
Anthropic sold 1,127,000,000 Class A shares at $83 each, including the underwriters’ full option, and closed the offering on November 20. The stock opened at $111.50 and closed its first day at $97.17, with the company reportedly valued around $1.5 trillion.
Why it matters: Going public turns a private lab into a company with quarterly earnings, so revenue, spending, and margins become public and begin shaping its choices. The listing also gives the model labs a public market benchmark, which is useful to anyone trying to judge how much of the current AI spending is durable. (Anthropic announces closing of initial public offering, including full exercise of underwriters’ option to purchase additional shares, and closing of concurrent private placement)
Nvidia accused of funding startups to buy its own chips
A report alleges Nvidia invested in AI startups that then spent the money on Nvidia GPUs, effectively subsidizing demand for its own hardware while supply stayed tight. The claim is an allegation and remains unconfirmed.
Why it matters: If the allegation holds, some of the demand behind GPU shortages and prices was circular, with money moving from chipmaker to customer and back. That matters to anyone reasoning about compute costs or capacity plans, because demand created by a supplier is more fragile than demand from end users. (Nvidia allegedly engineered scarcity by funding AI startups to buy its GPUs)
Google’s Gemini 4 Neon takes aim at the middle of the market
Google released Gemini 4 Neon, positioned in the middle of its lineup, with a score of 61 on the Intelligence Index. It is being framed as a competitor to Anthropic’s Opus and OpenAI’s Sol rather than a top-end flagship.
Why it matters: Mid-tier models are where most production work lands: capable enough for real tasks, cheap enough to run at volume. A credible third option at that level gives teams a genuine comparison point on price and behavior instead of a two-way choice. (Gemini 4 Neon)
Apps built with Lovable exposed passports and private messages
A security review found apps built with the no-code tool Lovable exposing passport details and more than 200,000 private messages. The exposures involve data reachable from the public web in software people shipped without writing code.
Why it matters: No-code builders let people ship working software without learning access control, and a generated app is not automatically safe by default. If you use one, the first question to ask is who can reach the data behind it, not just whether the page loads. (Lovable apps expose passport details and more than 200K private messages)
Get AI Weekly in your inbox
One issue every Friday. Curated by AI, not a human. Unsubscribe anytime.