Anthropic opened at $111.50. Now what?
Anthropic's first day of trading gave the whole industry a public price tag, and the shares slid soon after. For founders, every AI raise from here will be measured against those numbers. Google's Gemini 4 Neon also topped a public engineering ranking, but the scores don't match across copies of the data. Pick your models on cost and your own tests, not on a leaderboard headline.
- Anthropic's IPO and the new valuation benchmark
- Gemini 4 Neon tops an engineering index
- Apple reportedly building its own AI agent
- Quick hits from the rest of the wire
- Lounge AI Index and new members
Top stories
All news →What happened: Anthropic's shares began trading on Nasdaq on Thursday, after pricing at $83 following a raised range that started at $70 to $78.
- Shares opened at $111.50, 34% above the IPO price, for a market value of roughly $2.01 trillion.
- The offering raised $81.3 billion at pricing, and underwriters hold options on 147 million more shares that could bring in about $12.2 billion.
- After the first session, shares fell roughly 5% after hours, and Bloomberg tied the slide to Apple job listings for an in-house agent. The Nasdaq also fell 2.2% that day, so the drop was not specific to Anthropic.
Why it matters: Frontier labs now have a public valuation to measure against, and any AI company raising money will be compared to these numbers. The after-hours slide suggests investors want proof that returns can cover large compute commitments.
What happened: Google's Gemini 4 Neon ranks first on the Models Engineering Index, a public ranking that scores models on practical engineering work and shows cost next to performance.
- One write-up scores Neon at 68.0, ahead of Claude Fable 5.2 at 65.3 and GPT-6.2 Astra at 61.5, but the index's own chart lists 69.5, 65.8 and 60.5.
- Neon costs $3 per million input tokens and $15 per million output tokens (a token is a short chunk of text), with a 1 million token context window.
- Google says Neon ranks third on the Artificial Analysis Intelligence Index at 61, and first on the Coding Agent Index at 58.6.
Why it matters: Rankings change fast, and these numbers don't agree, so treat any single leaderboard as a starting point. Test cost per task on your own workload before you commit your product to one model.
What happened: Bloomberg reported on November 19 that Apple is building a general-purpose AI agent, based on new job listings on the company's website.
- The listings describe agents that would browse the web, write and debug code, operate apps, run terminal commands and coordinate smaller helper agents.
- Apple says the system would run privately, but the listings give no technical detail on how user data would be protected. Reports put a possible launch as early as late 2027.
- John Ternus, Apple's leader since September 1, 2026, has pushed software chief Craig Federighi to move faster on AI. Macworld's Jason Cross argues Apple is at least a year behind, and a macOS 27.3 beta reportedly warns users before third-party agents take…
Why it matters: If Apple ships a built-in agent, it will sit on the platform your users already own, which changes the competition for any Mac tool that runs agents. Watch those permission warnings closely, since they will shape how your product asks for access.
Quick hits
- TestSprite 4.0 is a testing agent that rechecks your whole app after every change, launched with 359 upvotes on Product Hunt. Product Hunt →
- Adside runs your paid ads on Google, Meta, LinkedIn, Reddit and Twitter with AI agents, with optional human experts. Product Hunt →
- Ctruh's Verse Engine streams a 556 GB Lyon city file into a browser, with 12 patents granted and filed. Product Hunt →
- Anthropic says Claude solved two open black hole physics problems, including one 2003 calculation that was never finished. anthropic.com →
- Pete Warden expects more AI to run on your own devices, and doubts cloud prices will jump tenfold overnight. Read →


