| ← All issues | |
| The Diff No. 047 Thursday, 19 November 2026 · AI dev news, only what is verified |
 |
Anthropic priced the biggest IPO on record last night - $83 a share, above its range, upsized to 980 million shares. It starts trading on Nasdaq today. Hours earlier, Google shipped Gemini 4 Neon, a frontier coder at mid-tier prices. One company just asked public markets for $81.3 billion to fund the next training runs, and got above the top of its range. The other answered with a model it says codes at the frontier for $3 a million input tokens. Both stories are about the same question: who pays for the next round of AI capacity, and at what price per token. Anthropic’s IPO prices the compute plan. Neon prices the work. Today you get a number for each. |
IN THIS ISSUE  | 01 🔔 Anthropic priced at $83 - above its range, upsized, and the biggest IPO on record. Trading starts today. 980,000,000 shares at $83.00 - more shares than announced, above the $70-78 range. NPR puts the raise at a record $81.3 billion. ANT debuts on Nasdaq today; the first print is still ahead. |  | 02 ⚡ Google shipped Gemini 4 Neon hours before the bell - frontier coding claims at $3 a million tokens. One million-token context, #1 on the Artificial Analysis Coding Agent Index per Google, rolling out in the API, AI Studio, Antigravity and the Gemini app for paid tiers. VentureBeat frames it as Opus 5.5-class coding at 25% lower prices. |
|
🔥 VIRAL The loud ones on Hacker News. Numbers are a morning snapshot. |  01 Anthropic’s official pricing announcement, Nov 18 · SEC registration effective Nov 18 · Nasdaq debut expected today under ANT 🔔 Anthropic priced at $83 - above its range, upsized, and the biggest IPO on record. Trading starts today.980,000,000 shares at $83.00 - more shares than announced, above the $70-78 range. NPR puts the raise at a record $81.3 billion. ANT debuts on Nasdaq today; the first print is still ahead. SCENE Yesterday morning this was a clock story: pricing due during US hours, above-range chatter from unnamed sources (No. 046). Overnight it became a fact. Anthropic published the pricing notice itself - no anonymity required. THE THING
The verified facts, from Anthropic’s own announcement: the IPO priced at $83.00 per share, above the marketed $70 to $78 range. The offering was upsized to 980,000,000 Class A shares - 956,351,354 sold by Anthropic and 23,648,646 by selling stockholders, whose proceeds do not go to the company - up from the 825,000,000 announced at roadshow launch. Underwriters hold a 30-day option on up to 147,000,000 more. The SEC declared the registration effective November 18; shares are expected to trade on the Nasdaq Global Select Market today, November 19, under the symbol ANT, with closing expected November 20. NPR reports the raise at a record $81.3 billion; AP describes it as the biggest Wall Street debut of all time at a roughly $1.5 trillion valuation. One thing that has not happened yet at our send time: the first trade. Every “debut” number you see before the Nasdaq opening cross is positioning, not a price. ✂️ STEAL IT Steal the dilution math. Anthropic sold roughly 956 million new shares; at $83 that is about $79 billion of fresh primary capital - money the company itself says funds the compute build-out. Yesterday Nvidia’s release said Anthropic’s Grace Blackwell and Vera Rubin access through cloud providers was expanded; today’s price is what that access costs, financed. If Claude is load-bearing in your stack, watch two numbers over the next weeks: where ANT settles after the first-day pop, and whether the underwriters exercise the 147-million-share greenshoe - both tell you whether public money will keep underwriting the capacity your roadmap assumes. Watch VERDICT
Nothing to patch or migrate; this is a capital-structure event, not a product one. The useful habit: separate the pricing fact (official, $83, 980 million shares) from the valuation framing (AP’s $1.5 trillion is arithmetic on reported share counts, not an Anthropic statement) and from today’s open, which has not happened at send time. This fails you if you read a strong debut as a verdict on Claude’s quality - it prices the compute plan, not the model. Anthropic: the official pricing announcement · NPR: record $81.3 billion raised · AP: the $1.5 trillion debut framing |
 02 Google’s official announcement, Nov 18 · model card live · benchmark rankings are Google’s, from its post ⚡ Google shipped Gemini 4 Neon hours before the bell - frontier coding claims at $3 a million tokens.One million-token context, #1 on the Artificial Analysis Coding Agent Index per Google, rolling out in the API, AI Studio, Antigravity and the Gemini app for paid tiers. VentureBeat frames it as Opus 5.5-class coding at 25% lower prices. SCENE While Anthropic’s bankers closed the book, Google DeepMind posted Neon - the third Gemini 4 model in roughly a week, after Carbon and Hydrogen. Google’s own framing: Carbon for the deepest long-horizon work, Hydrogen for rapid everyday execution, Neon in the middle. THE THING
The verified facts, from Google’s announcement and model card: pricing is $3 per million input tokens - $0.15 cached, a 95% discount - and $15 per million output, with a one million-token context window. Google says Neon ranks #3 overall with a score of 61 on the Artificial Analysis Intelligence Index at xhigh thinking, #1 overall on its Coding Agent Index at 58.6, and #2 on the Vals Index at 70.9%. Distribution is live now: the Gemini API in AI Studio, Gemini for business, Google Antigravity, and the Gemini app for AI Pro and Ultra subscribers - 9to5Google notes it replaces Argon in the picker, with 3.5 Flash-Lite renamed “Auto.” VentureBeat’s headline comparison - Opus 5.5-level coding at 25% lower API token prices - is VentureBeat’s framing, not Google’s claim, and “level” there is their read of the benchmark table, not an independent bake-off. ✂️ STEAL IT Steal the eval protocol, not the leaderboard. Neon’s pitch is price per completed task, so run the comparison that actually matters: pick one real, annoying task from your own repo - a migration, a flaky test hunt - and run it in AI Studio on Neon against your current default. Time to a passing result times token price is the only leaderboard that bills you. Note what Google chose to headline: cached input at a 95% discount. That is a bet that your workload is re-read-heavy - long context, agentic loops - which is exactly where coding agents spend. Try today VERDICT
If you already pay for a coding model, an afternoon in AI Studio costs you nothing but the task you bring. Treat every ranking above as Google’s account of its own model until third-party evals land; the AA and Vals numbers are quoted from its post. This fails you if you switch production traffic on launch-day benchmarks - run your task first. Google: the official Gemini 4 Neon announcement · Google DeepMind: the Neon model card · VentureBeat: the Opus 5.5 price framing (their comparison) |
📊 THE ATTENTION GAP HN points at about 06:30 IST. The biggest IPO in history, priced hours earlier, drew 11. The model launch drew 576. Builders vote for what they can run tonight. | Anthropic prices IPO at $83 thread | | 11 HN | | Gemini 4 Neon thread | | 576 HN |
|
ALSO ON THE RADAR Nvidia’s first all-hands without Huang: “Let’s move forward.” Interim CEO Colette Kress addressed employees Wednesday at the first company-wide meeting since the resignation, per CNBC, quoting a person in the room speaking anonymously. CNBC and Semafor both report the board and staff are split over calls to bring Huang back; AP reports the company rejected his offer to return despite pressure from Silicon Valley and Trump. All of it is outlet reporting - the only on-record board statement remains Monday’s. CNBC: the all-hands report → Tencent quietly open-released Hy4 flash: 224B total, 12B active, 1M context. Tencent’s November 16 announcement: Hy4 flash, its “most cost-effective” Hy model, with 224 billion total parameters but only 12 billion active per token, native text-and-image input, and a one million-token context window; weights and config are on Hugging Face under tencent/Hy4-flash. The cost per token of open models keeps sliding under the closed APIs. Tencent: the Hy4 flash announcement → Claude Meetings is still in monitoring, three days in. Anthropic’s status page still shows the November 17, 23:39 UTC update as the latest: joining calls more reliably, Enterprise enabled, Team customers adding manually, Auto-join restoring gradually. No full-restoration note yet, and no cause statements for any of the November 13-17 incidents, including both Apple sign-in failures. Anthropic status page → |
🧪 ONE THING TO TRY TONIGHT Ten minutes: give Neon one real task from your own backlog. Open Google AI Studio, select Gemini 4 Neon, and hand it one genuine task - the migration note you keep postponing, a failing test with a long log. Run the identical prompt against your current default model. Compare wall-clock to a usable answer, then glance at the token counts: at $3 in and $15 out, the question is not whether Neon tops a chart, it is whether it finishes your task cheaper. Bring the task; ignore the leaderboard. |
What this issue could not see. Anthropic’s price, share counts, greenshoe, dates and ticker are from its official pricing announcement; the record-raise figure is NPR’s reporting and the $1.5 trillion valuation is AP’s framing. Trading had not begun at our send time, so no opening price appears here. Gemini 4 Neon’s benchmark rankings, scores and pricing are Google’s statements about its own model, quoted from its announcement and model card; the Opus 5.5-level comparison is VentureBeat’s framing, not an independent test. HN points are a snapshot at about 06:30 IST. Nvidia all-hands and board-split details are anonymous sourcing via CNBC and Semafor; AP’s return-rejection report is attributed reporting, not a company statement. Hy4 flash specifications are Tencent’s, from its announcement and model card. No video this issue: no launch video from a verified official channel was found. Reddit and X were not consulted. |
| Images: credited in each report; source screenshots and announcement artwork, not our tests. |
|