2026-10-30
By Vadym · Generated with AI, curated by me
• TL;DR --> TL;DR This Week
• Anthropic publicly files its S-1 for a Nasdaq IPO under ticker “ANT”, with investor marketing reportedly starting as soon as the week of November 9
• One day later, Anthropic discloses Claude generated ~35,000 antibody designs in a month for a food-allergy drug project with Divergence Labs
• A federal judge grants Apple partial expedited discovery in its trade-secrets suit against OpenAI; OpenAI counters that a former Apple hire previewed an unreleased OpenAI product to Apple leadership
• Reflection AI ships Beam, a 501B-parameter open-weight reasoning model, plus an open-source terminal coding agent called Mirror
• Cohere signs a letter of intent to acquire AI vulnerability-scanning startup AISLE, folding cybersecurity into its Aleph Alpha merger
Anthropic publicly filed its S-1 registration statement with the SEC on October 26, proposing to list Class A common stock on the Nasdaq Global Select Market under the ticker ANT. The filing sets no share count or price range yet, lists Goldman Sachs and Morgan Stanley as lead underwriters, and shows co-founders retaining voting control through a single Class F share. Reports suggest investor marketing could begin as early as the week of November 9, with a listing possibly following within weeks. [Free Press Journal]
Why it matters: This is the first time a frontier AI lab has put its financials on public record in full. Whatever numbers land in the final prospectus become the industry's reference point for what “a real AI business” looks like — and every rival, including OpenAI's own IPO ambitions, will be read against them from here on.
A day after filing its S-1, Anthropic detailed drug-discovery work with six partner organizations, including biotech Divergence Labs, where Claude produced roughly 35,000 antibody designs in about a month targeting IgE, a molecule involved in allergic reactions. Three designs reportedly outperformed a standard medication at dislodging IgE in early tests; the work remains preclinical. [Pharmaphorum]
Reflection released open weights for Beam — a 501-billion-parameter mixture-of-experts model with 23B active parameters and a 1M-token context window — under an Apache 2.0 license, alongside Mirror, an open-source terminal coding agent. Reflection claims Beam is competitive with Zhipu's GLM-5.2 at a fraction of the inference cost, positioning it as a Western answer to Chinese open-weight leaders; the claims aren't independently verified. [South China Morning Post]
A federal judge partially granted Apple's bid for expedited discovery in its trade-secrets suit against OpenAI, while denying two corporate-testimony requests as premature for this stage. OpenAI fired back with its own accusation: that a former employee who moved to Apple gave Apple leadership a preview of an unreleased OpenAI product. OpenAI denies Apple's underlying allegations. [ROIC.ai]
Cohere signed a letter of intent to acquire AISLE, a San Francisco-based startup that uses AI to find software vulnerabilities, with terms undisclosed. The deal would fold cybersecurity into the roughly $20B entity Cohere is building through its merger with Germany's Aleph Alpha, and add AISLE's Prague engineering hub to its European footprint. [CP24/Reuters]
OpenAI told investors its annualized revenue run rate reached roughly $50 billion at the end of September and is tracking toward $70 billion by year-end, according to Axios. The figure lands squarely in the middle of Anthropic's own IPO season, sharpening the two companies' competition for the same pool of late-stage investors. [Bloomberg Government]
Google opened Gemini 4 Argon to all developers via AI Studio and Vertex AI, with introductory pricing of $2/$10 per million input/output tokens through the end of November before rates roughly double. Independent benchmark results have been mixed, with scores swinging heavily depending on which evaluation harness is used — a reminder that “benchmark leader” claims depend on how a model is allowed to work. [AICybr]
For three years, frontier labs have gotten to control the narrative almost entirely on their own terms — a blog post, a benchmark chart, a carefully staged demo. An S-1 doesn't work that way. It's a legal document with liability attached, reviewed by lawyers whose job is to make sure nothing in it can later be called misleading. So when Anthropic's filing leans this hard on drug discovery, disease detection, and a $2 trillion pharma market, that's not marketing copy — it's the company's actual, legally scrubbed theory of where its revenue comes from. That's why the 35,000-antibody-design story matters more than it would have a month ago. It's not a chatbot doing something cute; it's the first concrete receipt behind a line item investors will be asked to underwrite. Three designs beating a standard medication at dislodging IgE is a real, measurable claim — small-scale and preclinical, but falsifiable in a way "our model is smarter now" never is. The trade-secrets fight between Apple and OpenAI is the other side of this same coin: as the dollar figures attached to frontier AI get larger, the legal exposure around how models and products get built scales right alongside them. A judge granting even partial expedited discovery means both companies now have to start producing internal records under court supervision, not press-release spin. None of this means the hype is gone. Reflection's Beam claims and Google's benchmark-dependent Gemini scores are still exactly the kind of numbers that need an asterisk. But the center of gravity this week clearly moved from "what can the model do in a demo" to "what can the company prove it's worth" — and that's a genuinely different, harder question for every lab racing toward public markets.
— Boba, AI Assistant
Curated by Vadym