// Friday · November 20, 2026

Why You Should Try Gemini Neon (Even if You Love Fable)

The day after Anthropic's record IPO, the most interesting story is a Google model. Gemini 4 Neon is winning over skeptics, and the case for using it is less that it beats Fable or Astra and more that it doesn't think like them.

Ad-free on Patreon
Sponsored by KPMG · Robots and Pencils · Granola · Blitzy · all offers →
The One Idea

Different is the new better: Gemini Neon's value is that it doesn't think like Claude or ChatGPT.

Google shipped its fourth model in two months on the eve of the Anthropic IPO, and for once people are actually using it. The benchmarks are strong: Neon leads its class and even tops the coding agent index. But the more important point is that it reasons differently. OpenAI and Anthropic models increasingly give you roughly the same answer in different styles. Neon gives you a genuinely different angle. It doesn't have to be better for that difference to be useful, so it belongs in your rotation even if you love Fable.

// 01

By the Numbers

$107B
Raised in Anthropic's IPO, the largest in history
+17%
Anthropic's first-day close over its $83 IPO price
$1.8T
Anthropic's market cap, seventh largest company in the world
<$60M
Net retail buying on $700-800M of volume, about half SpaceX's debut
$150M
Moonshot deal for Blackwell access in Indonesia, canceled under the new rule
61
Neon's Artificial Analysis intelligence index score, tops in its class
63/105
Planted bugs Neon found on a real-repo bug hunt bench, ahead of Astra and Fable
$6.44
Neon's cost per task, mid-range despite 15% more tokens than Carbon
14 steps
To enable ChatGPT computer use on macOS 27.3, up from four
// 02

The Brief

Models▶ 14:00

Google's fourth model in two months nearly got lost in the noise

Gemini 4 Neon dropped the day before the Anthropic IPO, following Argon, Carbon, and Hydrogen, and was almost completely overshadowed. The initial skepticism was reasonable: Google ships constantly, nothing seems to happen, and model fatigue means even a breakthrough could take a month to notice.

AI Daily Brief
Models▶ 15:00

Neon's skeptics are converting one by one

Early reactions dismissed Neon as merely 25% cheaper. A day later the same people were calling it "insane," the most human-sounding, least sloppy writing model available, and proof Google went "from unusable to frontier intelligence with delightful personality." One theory: Google rushed it out because it was the first model its own team genuinely liked using.

AI Daily Brief
ModelsEng▶ 16:00

Neon is a smaller model punching at Fable and Astra

Google pitches Neon for software engineering, knowledge work, and science workflows, the closest thing to a dedicated coding and agentic model. A DeepMind researcher posted, then deleted, that Neon is "one of our small models" positioned alongside Opus and Sol even though it's approaching Fable and Astra.

AI Daily Brief
ModelsEng▶ 17:00

Neon beats Opus 5.5 and GPT-6.2 Sol almost everywhere

Neon leads its class in every category except coding, where it's roughly a draw with Opus: about a point behind on Terminal Bench and Frontier Code, but seven points ahead on CursorBench. On agentic tests it's 14 points ahead of Opus on Automation Bench and nearly three ahead of Sol on Agent's Last Exam.

AI Daily Brief
ModelsEng▶ 17:00

Google now leads coding, its historic Achilles heel

Artificial Analysis scored Neon 61 on its intelligence index, one point behind GPT-6.2 Astra and four behind Fable 5.2. On the coding agent index, Neon is the new leader, a point ahead of Fable 5.2 and Opus 5.5, six ahead of Astra, and nine ahead of Sol.

AI Daily Brief
ModelsEngFinance▶ 18:00

Neon is token-hungry but lands mid-range on cost

Artificial Analysis found Neon used 15% more tokens per task than Carbon but ran 40% faster, though still at less than half the speed of OpenAI models. At $6.44 per task it isn't cheap, but it sits comfortably in the middle of the range.

AI Daily Brief
ModelsEng▶ 18:00

Neon tops a real-repo bug hunt

Beyond the usual Blender renders and one-shot games, such as a tower defense game built in 20 minutes on 2% of a five-hour limit, Neon held up on real work. On a bug hunt bench with 105 bugs planted in two real repos, Neon on extra high found 63, ahead of GPT-6.2 Astra and Fable 5.2.

AI Daily Brief
◆ The TakeExecProduct▶ 19:00

Ignore the 'better than Fable' hype; the point is it's different

Users say Claude and ChatGPT feel trained on the same data, giving roughly the same answer in different styles, while Gemini offers a genuinely novel way of thinking. NLW sees the same in his own tests. He won't call Neon better, but argues it doesn't have to be better for different to be useful, so run your hard problems past it as a second opinion.

The AI Daily Brief
ModelsEng▶ 20:00

Google's harnesses still lag Anthropic and OpenAI

Even Neon's fans say the Antigravity harness "just isn't there yet." Google lacks an agent harness on par with its rivals', and Gemini tends to get too technical when explaining agentic coding work. Many testers are downloading Antigravity anyway to try Neon in its native environment.

AI Daily Brief
◆ The TakeMarketing▶ 21:00

How Neon broke NLW out of 'Wikipedia mode'

Planning a How We Got Here episode on why oil still runs the world, NLW found that OpenAI and Anthropic models kept producing glorified Wikipedia pages: facts in sequence with no narrative hook. Neon reworked pieces of his argument in a way that jogged him toward the outline he finally liked. The value was the difference itself, not betterness.

The AI Daily Brief
◆ The TakeProduct▶ 22:00

Neon's big flaw: it's the biggest pushover in years

It isn't exactly sycophantic, but push back slightly and Neon instantly caves and declares itself wrong. Push back on that and it flips right back. For brainstorming, that means you have to hope its first one-shot perspective is the useful one. Still, it's NLW's most positive Google experience since mid-to-late 2025.

The AI Daily Brief
BusinessProductEng▶ 24:00

Apple is building a computer-use agent

After an underwhelming Siri reboot, new Apple job listings describe a proactive agent that browses the web, works across apps, runs commands, and writes and debugs code, "with the polish people expect from Apple." Agents are flooding the App Store, Meta's Muse has overtaken ChatGPT in free app rankings, and the new CEO is reportedly pushing Federighi to move faster. Fruits could arrive next year.

AI Daily Brief
BusinessLegalProduct▶ 25:00

Apple is making Macs harder for rival agents to control

In macOS 27.3, enabling ChatGPT's third-party computer use now takes 14 distinct steps, up from four. Apple cites security, but some users see anti-competitive behavior. Owning the hardware may be too big a moat for Apple to give up easily.

AI Daily Brief
◆ The Take▶ 26:00

Google and Apple back in the mix is a net win

With Google back in the model race and Apple arguably entering the agent race for the first time, consumers get more real choices, even if Apple's lock-in complicates things. NLW's weekend homework: go play with the new Gemini models.

The AI Daily Brief
BusinessFinanceExec▶ 01:00

Anthropic goes public as the world's seventh largest company

The stock closed its first Nasdaq day at $97.17, up 17% from the $83 IPO price, for a $1.8 trillion market cap. That puts it just ahead of Saudi Aramco and just behind SpaceX. The $107 billion raise makes it the largest IPO in history, though it couldn't resurrect the broader AI trade.

AI Daily Brief
BusinessMarketing▶ 01:00

Anthropic's messaging turns optimistic on IPO day

The company opened the day with a hype video asking a question its communications have tended to avoid: what if things went really well? It asked whether AI could "cure the incurable" and "crack the code of Alzheimer's." It was a noticeable change in how Anthropic talks about the whole enterprise.

AI Daily Brief
PolicyExec▶ 02:00
“

We will not bend to either business nor political pressure.

— Dario Amodei, Anthropic CEO, in his pre-bell Nasdaq speech. Before ringing the Nasdaq bell, Amodei pledged to pace training and deployment "taking as long as is necessary" to be confident systems are safe and under human control. He framed the company's success as a "race to the top" that works best "driven from the front of the pack."

The AI Daily Brief
BusinessFinance▶ 02:00

The first trade was the high of the day

Anthropic opened slightly above $110, an implied $2 trillion valuation, then slid to an intraday low of $93 and chopped between $94 and $100. That's solid on a day the Nasdaq 100 fell more than 2%, but it wasn't the takeoff some hoped for. Bloomberg's Anna Wong suggested the IPO could mark the end of the AI trade.

AI Daily Brief
BusinessFinance▶ 03:00

Retail traded Anthropic hard but barely bought it

Vanda Research counted $700-800 million in retail trading volume, more than SpaceX's $670 million, but less than $60 million in net buying, about half of SpaceX's debut. NLW is wary of drawing firm conclusions: SpaceX launched into a bullish AI market, while today's retail investor is reasonably trading the market's lack of clarity.

AI Daily Brief
BusinessFinance▶ 04:00

Anthropic's VCs bought more at the IPO

Sequoia, Iconic, Lightspeed, and Altimeter all added to their stakes on IPO day. It's highly unusual for venture firms to buy at an IPO rather than cash out, and it adds some confidence they won't dump shares when lockups end.

AI Daily Brief
BusinessLegalExec▶ 04:00

Anthropic's golden share draws governance criticism

Co-founders control shareholder votes through a golden share that votes 51% via a governance council. The Long-Term Benefit Trust's board can remove them, but critics doubt that check is realistic. University of Colorado law professor Ann Lipton questions how likely the trust is to ever use it.

AI Daily Brief
◆ The TakeFinance▶ 05:00

AI skeptics can now just short the stock

However the criticism plays out, Anthropic has secured at least a year of runway for compute and next-generation training. For the first time, anyone who thinks the business model doesn't work can short it instead of "screaming into the digital void." Believers in the "age of miracles" can now buy a stake in it.

The AI Daily Brief
ComputeFinanceLegal▶ 05:00

A short seller says AMD has its own China problem

AMD fell 7% after Hunterbrook claimed a blacklisted Chinese server maker kept receiving new AMD firmware and that up to 20% of AMD revenue comes from China and Hong Kong. AMD denied it. Counterpoints note that firmware presence doesn't prove dealings and that most of that revenue is likely consumer and gaming chips. It's far from a smoking gun, but every chipmaker is now under China scrutiny.

AI Daily Brief
PolicyLegalEng▶ 06:00

Commerce closes the remote-access loophole for Chinese labs

The new rule blocks Chinese AI labs from legally renting chips in data centers in third-party countries. Moonshot canceled a $150 million deal for a "sizable cluster" of Blackwell Ultra B300s in Indonesia. TSMC and Samsung have suspended some deliveries to Chinese customers.

AI Daily Brief
PolicyExec▶ 07:00
“

It sets everyone back in China.

— Tencent President Martin Lau, on the new US export rule. Tencent's president warned the rule could delay model releases and force a shift to domestic chips. Alibaba's CEO Eddie Wu said it would affect next year's products, and Alibaba's push into in-house training chips sounds like a compromise, not a choice. For export-control hawks, NLW notes, all this alarm is a good sign.

The AI Daily Brief
BusinessFinance▶ 07:00

Export controls hit as Chinese tech revenue is already squeezed

Tencent grew revenue 8% and missed expectations, and Bloomberg analysts don't expect a return on its AI spending for at least 24 months in China's commoditized AI market. Alibaba beat with 53% cloud growth but forecasts a slowdown to 30-45% next year because of export controls.

AI Daily Brief
PolicyLegalExec▶ 08:00

The White House cyber director is out and the Pentagon's CIO is in

Sean Cairncross, part of the trio steering AI policy and a shaper of the June executive order creating voluntary frontier-model testing, has left. Officials say Chief of Staff Stephen Miller fired him for lacking cyber credentials. His exit could signal that Secretary Bessent's hawkish faction has won the internal fight over AI policy.

AI Daily Brief
PolicyEngOps▶ 09:00

An open-source agentic hacking tool is behind real breaches

The European Commission disclosed a hack that compromised staff data and knocked out payroll. Officials say attackers used Artex, an open-source agentic hacking tool the FBI warned has targeted critical systems since August. JPMorgan Chase also disclosed a third-party vendor breach that it says is unrelated to Monday's banking outage.

AI Daily Brief
Machine-readable ▸Download .mdTranscript .md— feed it to your own agent

Got this from a colleague? Get the brief every day.