Anthropic Spent $10M Roasting OpenAI on National TV. Then Both Companies Dropped Flagship Models 20 Minutes Apart.

The week of February 3-8, 2026, will go down as the most chaotic seven days in the history of AI.

On Tuesday, Anthropic dropped a Super Bowl ad campaign openly mocking OpenAI for putting ads in ChatGPT. On Wednesday, Sam Altman posted a 400-word tirade calling Anthropic “authoritarian” and “dishonest.” On Thursday, both companies released flagship AI models within 20 minutes of each other — triggering a near-trillion-dollar stock selloff. And on Sunday, their competing ads aired during the most-watched broadcast in America.

This is the full story of how two AI companies went to war on every front simultaneously — and what actually happened when the dust settled.

The Ad That Broke Sam Altman

On February 4, Anthropic dropped its first-ever television campaign: four spots created by agency Mother and directed by Jeff Low via Biscuit Filmworks. The tagline: “Ads are coming to AI. But not to Claude.”

Each spot follows the same structure: someone asks an AI chatbot a deeply personal question, gets helpful advice for a few seconds, then watches in horror as the response pivots into an aggressive product pitch. A man asking a therapist-bot how to communicate better with his mother gets redirected to “Golden Encounters, the mature dating site that connects sensitive cubs with roaring cougars.” A skinny guy doing pull-ups asking about six-pack abs gets pitched “StepBoost Max” height-increasing insoles. A nervous entrepreneur seeking business advice gets hit with a payday loan plug — “because girlbosses need SHE-E-O money quick.”

Set to Dr. Dre’s “What’s the Difference,” the ads were funny, sharp, and — most importantly — timed to land three weeks after OpenAI announced it would begin testing ads in ChatGPT’s free tier.

Felix Richter, Chief Creative Officer at Mother, explained the concept: “People asking AI about their health, their relationships, their business. Then a sponsored answer. We don’t need to explain why that’s wrong. We just need to show it.”

The timing was surgical. The target was obvious. And Sam Altman completely took the bait.

”I Laughed. But.”

Within hours of the ads dropping, Altman posted a lengthy response on X that started with a compliment and rapidly devolved into something much more revealing. Here it is, pulled from the original post and corroborated across TechCrunch, Variety, and Ad Age:

“First, the good part of the Anthropic ads: they are funny, and I laughed. But I wonder why Anthropic would go for something so clearly dishonest. Our most important principle for ads says that we won’t do exactly this; we would obviously never run ads in the way Anthropic depicts them. We are not stupid and we know our users would reject that.”

Then it escalated:

“I guess it’s on brand for Anthropic doublespeak to use a deceptive ad to critique theoretical deceptive ads that aren’t real, but a Super Bowl ad is not where I would expect it.”

Then it went further:

“Anthropic serves an expensive product to rich people. We are glad they do that and we are doing that too, but we also feel strongly that we need to bring AI to billions of people who can’t pay for subscriptions.”

“More Texans use ChatGPT for free than total people use Claude in the U.S.”

Then it went nuclear:

“Anthropic wants to control what people do with AI — they block companies they don’t like from using their coding product (including us), they want to write the rules themselves for what people can and can’t use AI for, and now they also want to tell other companies what their business models can be.”

“One authoritarian company won’t get us there on their own, to say nothing of the other obvious risks. It is a dark path.”

For context: Anthropic was founded in 2021 by former OpenAI researchers — including siblings Dario and Daniela Amodei — who left over disagreements about the company’s direction on safety. Altman calling them “authoritarian” is deeply personal.

OpenAI CMO Kate Rouch piled on in her own X thread, jabbing that the ads were made by “the former Meta ads executive” — a reference to Anthropic’s head of brand marketing Andrew Stirk, who previously worked at Meta. She wrote: “Real betrayal isn’t ads. It’s control. Anthropic thinks powerful AI should be tightly controlled in small rooms in San Francisco and Davos. That it’s too DANGEROUS for you.”

The Internet’s Verdict: “You Should Have Just Rolled With It”

The media response was swift and almost uniformly brutal — for Altman.

Futurism called his response “an unintentional masterclass in corporate insecurity,” noting that “it’s always a bad sign when someone insists that they’re not mad and actually laughing.”

Inc. ran the headline: “Anthropic’s ‘Dishonest’ Ads Clearly Struck a Nerve With Sam Altman. That Was the Point.”

PC Gamer — whose URL slug read “openai-sam-altman-outs-himself-as-massive-baby” — noted that Altman “claimed to find the ads funny, but then unleashed a further 400 words on why the ads are ‘not funny, actually.’”

The SF Standard asked simply: “Can OpenAI take a joke?” Of Rouch’s “Real betrayal isn’t ads. It’s control” line, they added that it “kind of sounds like something AI would say.”

Entrepreneur described it as a “novella-long rant.” Fast Company compared the whole spectacle to the Pepsi vs. Coca-Cola wars.

The top-liked reply to Altman’s post on X: “You should have just rolled with it. Instead, it’s cope.”

Anthropic, for their part, barely engaged. Anthropic President Daniela Amodei told ABC News: “This really isn’t intended to be about any other company other than us.” She added: “People are sometimes uploading private or confidential information to their AI tool, and to us, it just didn’t feel like the respectful way to treat our users’ data.”

The official corporate statement was a masterclass in letting someone else dig their own hole: “Our business model is straightforward: we generate revenue through enterprise contracts and paid subscriptions. We respect that other AI companies might reasonably reach different conclusions.”

That was it. No counterpunch. No lengthy thread. Just a shrug.

Then They Dropped the Models

The ad war was theater. What happened on February 5 was the real battle.

At approximately 10:00 AM Pacific, Anthropic released Claude Opus 4.6 — its most powerful model ever. Twenty to thirty minutes later, OpenAI fired back with GPT-5.3-Codex. The timing was too precise to be coincidental. One developer-focused outlet called it “a distributed denial of developer attention.”

Reports suggest Anthropic actually pushed their announcement up by 15 minutes from the planned time, which may have caught OpenAI slightly off-guard. But OpenAI had pre-released its Codex app on February 2, meaning the model upgrade was already staged and ready to deploy.

Claude Opus 4.6: The Depth Machine

Anthropic CEO Dario Amodei framed it simply: “We’re upgrading our smartest model. Across agentic coding, computer use, tool use, search, and finance, Opus 4.6 is an industry-leading model, often by wide margin.”

The highlights:

1 million token context window (beta) — the first Opus model to support this. That’s roughly 700,000 words of context in a single conversation.
128K max output tokens — doubled from the previous 64K limit.
Adaptive thinking — Claude dynamically decides when and how much to reason, replacing the older manual budget_tokens approach.
Compaction API (beta) — automatic context summarization enabling effectively infinite-length conversations.
Fast mode — up to 2.5x faster output at premium pricing.

The benchmarks told a specific story — Opus 4.6 dominates reasoning-heavy, long-context, and professional tasks:

Benchmark	Score	What It Measures
ARC-AGI 2	68.8% (up from 37.6%)	General reasoning
MRCR v2	76% (vs. 18.5% for Sonnet 4.5)	Long-context retrieval
BigLaw Bench	90.2%	Legal/compliance
OSWorld-Verified	72.7%	Real-world computer use
GDPval-AA	1606 Elo	Professional task performance
Terminal-Bench 2.0	65.4%	Agentic coding

That last number matters because of what happened twenty minutes later.

GPT-5.3-Codex: The Speed Machine

Sam Altman called it “a very big step forward” and posted on X that “it feels like more of a step forward than the benchmarks suggest.”

The highlights:

25% faster than GPT-5.2-Codex, using less than half the tokens for the same tasks.
“First model that was instrumental in creating itself” — OpenAI used early versions of 5.3-Codex to debug training, manage deployment, and diagnose test results during its own development.
Interactive mid-task collaboration — users can steer GPT-5.3-Codex while it’s working without losing context. As Altman put it: “If you see a co-worker making a mistake and you don’t interrupt them, that’s rude, right? Like it’s deeply inefficient.”
Highest cybersecurity risk rating — OpenAI’s first model to hit “high” on their internal preparedness framework, with $10M in API credits offered for cybersecurity defense applications.

And one benchmark that immediately grabbed headlines:

Benchmark	GPT-5.3-Codex	Claude Opus 4.6
Terminal-Bench 2.0	77.3%	65.4%

GPT-5.3-Codex blew past Opus 4.6 on Terminal-Bench — the benchmark Anthropic had just claimed the top score on, approximately 30 minutes earlier. As a competitive move, it was devastating in its precision.

Alex Embiricos, OpenAI’s Codex Product Lead: “If you look at where we were and the amount of progress we’ve had in the past year on our models’ coding ability, it’s crazy. I think the velocity there has been super high.”

Ed Bayes, OpenAI’s Product Design Lead, captured the internal culture shift: “I spend 90% of my time in code now. A year ago, it was flipped, and it was maybe 10%.”

Altman also revealed a detail that raised eyebrows: “It was amazing to watch how much faster we were able to ship 5.3-Codex by using 5.3-Codex.” The model literally helped build itself.

Who Actually Won?

The honest answer: it depends on what you’re doing.

Benchmark	Claude Opus 4.6	GPT-5.3-Codex	Winner
Terminal-Bench 2.0 (agentic coding)	65.4%	77.3%	GPT-5.3-Codex
SWE-bench Verified	~79-81%	~80%	Tie
OSWorld (real-world computer use)	72.7%	64.7%	Claude Opus 4.6
ARC-AGI 2 (reasoning)	68.8%	—	Claude Opus 4.6
BigLaw Bench (legal)	90.2%	—	Claude Opus 4.6
Inference Speed	Baseline	~25% faster	GPT-5.3-Codex
Token Efficiency	Baseline	2x fewer tokens	GPT-5.3-Codex

Opus 4.6 wins on depth: reasoning, long-context, professional/enterprise tasks, and real-world computer use. GPT-5.3-Codex wins on speed: faster inference, fewer tokens, and terminal-based coding benchmarks. Both companies can credibly claim leadership depending on which numbers they highlight.

Developer reaction was split. One Reddit user captured the common sentiment: “Codex imo is far better. Opus is only good when you give it a big issue to solve. Codex with a single problem is far better imo.” Others pushed back on the pricing gap — Opus costs significantly more — arguing that “if they start losing pro customers because their pricing is 4x for no significant better performance, they might get in big trouble later.”

OpenAI reported the Codex app reached 500,000 downloads within three days. One developer posting on 36kr summed up the week: “I simply can’t keep up with the iteration speed of AI.”

Then the Market Crashed

The dual model releases didn’t land in a vacuum. They landed on top of Claude Cowork’s plugin launch from January 30 — which had already spooked the software sector — and the combined impact was catastrophic.

Over seven days, nearly $1 trillion was wiped from software and services stocks in what analysts began calling the “SaaSpocalypse”:

Thomson Reuters plunged 16% in a single day — its worst drop on record
Intuit fell 11%, bringing its year-to-date loss to 34%
S&P Global dropped 11%
ServiceNow fell 7.6%
Microsoft sank 5%
FactSet dropped 10%
Oracle fell 56% from peak

The Goldman Sachs basket of U.S. software stocks lost 6% in a single session — the worst day since April. The WisdomTree Cloud Computing Fund was down over 20% year-to-date.

Michael O’Rourke, Chief Market Strategist at Jonestrading: “For two years, we have been talking about how AI is going to change the world. In the past few weeks, we have seen signs of it in practice.”

Dan Ives at Wedbush pushed back, calling it an “Armageddon scenario for the sector that is far from reality.” Gartner analysts agreed the panic was premature but conceded the models “expose how much day-to-day knowledge work remains manual, making it ripe for automation.”

OpenAI’s Ad: “Builders”

Against this backdrop, OpenAI aired its own Super Bowl commercial — a 60-second first-quarter spot for Codex.

The ad follows a single through-line of curiosity — from a child’s hand tracing a cobweb, through classroom notebooks and science books, to lines of early code, a Linux DVD slid into an aging computer, and eventually to modern scenes of building alongside ChatGPT and Codex. The tagline: “We build the tools. You build the future.”

It was a deliberate course correction from last year’s abstract “Intelligence Age” spot — a $14 million pointillist art piece that OpenAI internally acknowledged underperformed. This year’s ad was grounded, human, and concrete. National spots ran alongside regional “Real Stories” mini-documentaries highlighting everyday people using ChatGPT to run businesses.

AdWeek’s verdict: “OpenAI Wants You to Build But It’s Unclear What.”

The ad was fine. Professional. Competent. But it had the misfortune of airing during the same broadcast as Anthropic’s campaign — which was still generating headlines, feuds, and memes days after dropping. In the battle for cultural attention, “competent” doesn’t beat “controversial.”

What This Week Really Means

Zoom out and the week reveals a clear strategic split between these two companies.

OpenAI’s bet: Scale and access. 300M+ weekly ChatGPT users, free tier with ads, the Codex brand push, a Super Bowl ad about everyday builders. Altman’s core argument — that AI should be free for everyone, funded by advertising if necessary — is a coherent philosophy. It’s also the same one that turned social media into an attention economy.

Anthropic’s bet: Premium positioning and enterprise revenue. $9 billion annualized run rate, 80%+ from enterprise customers, ad-free product, and a Super Bowl campaign designed to make the distinction feel existential. Dario Amodei told CNBC that Anthropic doesn’t “need to maximize engagement for a billion free users because we’re in some death race with some other large player.”

Both companies are reportedly planning IPOs by end of 2026. The Super Bowl was a pitch deck disguised as entertainment.

AdWeek noted the obvious parallel: “The real test won’t be whether the ad resonated creatively, but whether Anthropic can actually remain ad-free as monetization pressure mounts.” They drew a comparison to Netflix, which once promised it would never run ads.

Slate offered the most cynical read: the AI ad saturation at Super Bowl LX has “sector-specific recession-indicator vibes that have defined Super Bowls of the past” — comparing it to the cryptocurrency ads of Super Bowl 2022 and the dot-com ads of 2000.

The Bottom Line

Here’s the scorecard for the week:

Anthropic won the ad war. The Anthropic spots generated more coverage, more conversation, and more cultural impact than OpenAI’s “Builders” campaign by a wide margin. Altman’s reaction amplified the story tenfold. Every headline about the feud was a headline about Claude.

The model war was a draw. GPT-5.3-Codex wins on speed and terminal coding benchmarks. Claude Opus 4.6 wins on depth, reasoning, and long-context tasks. Both are genuine leaps forward. Developers are split. Neither company can credibly claim total superiority.

The market didn’t care who won. It just sold everything.

And somewhere in the middle of all this, 120 million Americans watched a skinny guy get pitched height-increasing insoles by his AI personal trainer, and thought: yeah, that does sound about right.

Sources: