04 of 05 · Daily Top Five

Today’s Top Five: Nov 19, 2026

By Willa Ravenwire · · 6 min

WIRE

Why it matters. Anthropic prices its Nasdaq IPO at $83 a share, above its range, as Google ships Gemini 4 Neon and Genie 4 and Nvidia RTX Spark desktops…

Anthropic goes public on Nasdaq this morning after setting a price above its own range, and Google spent Nov 18 shipping a new mid-tier Gemini model and a more interactive world generator. Below that, a small lab showed what focused reinforcement learning can do for an open 27B model, and Nvidia’s compact desktops for local models finally went on sale.

1. Anthropic Prices Its IPO at $83, Above the Range It Marketed

What happened. Anthropic priced 980 million Class A shares at $83 each on Nov 18, with trading on the Nasdaq Global Select Market under the ticker ANT set to begin Nov 19 and the deal expected to close Nov 20. That is a bigger and pricier deal than the 825 million shares at $70 to $78 the company marketed when its roadshow opened on Nov 10. Of the 980 million shares, about 956 million are new stock sold by Anthropic and about 23.6 million come from existing holders, and underwriters can buy up to 147 million more within 30 days.

The hype check. At $83 a share the full offering works out to roughly $81.3 billion before underwriting fees, but Anthropic’s own announcement gives no valuation, so the trillion dollar figures circulating this morning come from outside calculations, not the company. Pricing above the range shows strong institutional demand for the shares, which is a separate question from whether the business will grow into that price once it reports results as a public company.

Verdict: Supported. The price, share count and listing details come straight from Anthropic’s announcement, and they build on the roadshow terms we flagged in yesterday’s watch list.

2. Google’s Gemini 4 Neon Aims for Frontier Work at Lower Cost

What happened. Google introduced Gemini 4 Neon, a model with a 1 million token context window that sits between the heavier Carbon and the faster Hydrogen in the Gemini 4 family. It is available on the paid Gemini API tier, in Gemini for business and Antigravity, and to Google AI Pro and Ultra subscribers in the Gemini app, priced at $3 per million input tokens and $15 per million output tokens. Google says it ranks first on several agent and coding leaderboards and costs about a third less per task than Carbon at the high thinking setting.

The hype check. Nearly every number in the launch is reported by Google or by evaluators it picked, and some safety figures come from internal testing only. Google itself concedes Neon trails larger, slower models on the Artificial Analysis overall index and still points customers to Carbon for the hardest long jobs, so the real pitch is price for performance rather than a new top model.

Verdict: Promising but unverified. Availability and pricing are confirmed, but the benchmark lead needs independent runs before it counts.

3. Genie 4 Lets Ultra Subscribers Rewrite a World Mid Session

What happened. Google DeepMind and Google Labs made Genie 4 available in Project Genie for eligible Google AI Ultra subscribers, with a waitlist for Pro members. The world model lets users change a generated environment while exploring it by typing new prompts, and it adds live audio such as ambient sound and character dialogue. The older Genie 3 model also got longer sessions of 3 minutes, up from 60 seconds, plus the ability to link worlds together and start from 360 degree panoramas.

The hype check. Google gives no session length, resolution or regional list for Genie 4 itself, and it does not define who counts as an eligible subscriber. The company also lists camera latency, complex physical interactions, audio precision and long term scene permanence as unsolved, which are exactly the problems that separate a striking demo from a usable tool for games or training simulations.

Verdict: Promising but unverified. The release is real, but the “most realistic” claim is Google’s own, and the key limits are not public.

4. RadixArk’s DaisyRose-27B Shows What Targeted RL Buys an Open Model

What happened. New lab RadixArk released DaisyRose-27B, a reinforcement learning tune of Qwen3.8-27B published under Apache 2.0 with weights in BF16, FP8 and NVFP4. The team trained on 1,200 long agent tasks, each built around a known weakness of the base model, over 18 days on 144 H200 GPUs using its open source Miles framework. The biggest reported jump is on Terminal-Bench 4.1, where the score rises from 5.6 to 12.0, with smaller gains on coding and instruction following tests.

The hype check. The results are self reported, and most gains outside Terminal-Bench 4.1 are a few points on benchmarks where the base model was already strong. The more durable contribution may be the training recipe and the claim that its backend roughly doubled hardware utilization over Megatron, which other teams can test for themselves because the tools are open.

Verdict: Promising but unverified. The weights and license are public, so builders can check the numbers directly.

5. Nvidia’s RTX Spark Desktops Go on Sale for Local Models

What happened. Nvidia said compact RTX Spark desktops from ASUS, Dell and MSI are available now, with Acer, GIGABYTE, HP and a Microsoft Surface developer box to follow. The Windows machines pair a 20 core Grace CPU with a Blackwell RTX GPU and up to 128GB of unified memory, which Nvidia says is enough to run 120 billion parameter language models locally. The software bundle includes CUDA, Linux tools through WSL, agent blueprints and Unsloth for fine tuning.

The hype check. Nvidia’s post lists no prices, and “up to” applies to both the memory and the 1 petaflop performance figure, so lower configurations may not fit the largest models. Running a 120B model locally also says little about speed, and no independent tokens per second numbers have been published yet.

Verdict: Needs follow-up. The machines exist, but pricing and real world throughput will decide whether they beat renting cloud GPUs.

Also on our radar

  • Researchers from Duke, UIUC and Harvard posted Energy-Based Reasoners, a looped reasoning design that hits 99.9% on Sudoku-Extreme with 345 times less inference compute than the next best model, though every test so far is an algorithmic puzzle.
  • Google added temporary threads to AI Mode in Search, sessions that skip your history and personalization, rolling out globally on desktop and mobile browsers.
  • The Gemini app now runs on lower memory Android Go phones after Google rebuilt its architecture, bringing Gemini Live and image generation to devices that previously got only basic voice commands.

What to watch

The first trading days for ANT will show whether public investors match the demand that pushed Anthropic’s price above its range, and the company’s first quarterly report as a public firm will be the earliest hard look at its revenue and spending. On the model side, watch for independent evaluations of Gemini 4 Neon, since its case rests on matching top models at a lower cost per task.

Share on X

Sources

  1. Anthropic, Anthropic prices initial public offering
  2. Anthropic, Anthropic launches roadshow for proposed initial public offering
  3. Google, Introducing Gemini 4 Neon
  4. Google, Genie 4 is now available
  5. RadixArk on Hugging Face, DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency
  6. NVIDIA, NVIDIA RTX Spark Compact Desktops Out Now
  7. Chen, Gladstone, Li, Ji and Du, Energy-Based Reasoners (project page)
  8. Google, Temporary threads are now available in AI Mode in Search
  9. Google, The Gemini app is now available on lower-memory Android devices

By Willa Ravenwire · The AI Bandwagon

Next on today’s wire: 05 of 05Today’s Top Five: Nov 18, 2026

More in Daily Top Five

  1. Daily Top Five · 6 min

    Today’s Top Five: Nov 22, 2026

    The Lean and Mathlib communities say Alpöge's machine checked proof of the Riemann hypothesis is correct, plus a canary study of leaky inference providers.

    Nov 22, 2026

  2. Daily Top Five · 6 min

    Today’s Top Five: Nov 21, 2026

    Anthropic closed its IPO with about $93.5 billion raised, as two rival Riemann hypothesis proofs and a new open model from Thinking Machines landed.

    Nov 21, 2026

  3. Daily Top Five · 6 min

    Today’s Top Five: Nov 20, 2026

    Anthropic opens on Nasdaq at $111.50, Claude takes on black hole physics, and Gemini learns to drive your Mac.

    Nov 20, 2026