Anthropic's Super Bowl shot at OpenAI

Anthropic targets OpenAI with bold Super Bowl ads, GPT-4o faces retirement amid user backlash, GPT-5.3-Codex and Claude Opus 4.6 go head-to-head, and Google Gemini 3 surges past 8 million paying enterprise licenses.

william murray

08 Feb 2026 • 5 min read

From The Bit Baker Daily Briefing — February 8, 2026

PLUS: GPT-4o's emotional farewell, dueling model drops, and Google's AI surge

Good morning, Dave.

The AI giants aren't just competing anymore — they're throwing elbows. Dueling model drops, a Super Bowl ad war, and Google quietly pulling ahead in enterprise adoption have turned today's news cycle into something closer to a cage match.

In today's Bit Baker:

Anthropic Takes a Super Bowl-Sized Swing at OpenAI — Anthropic is running Super Bowl LX ads that directly mock OpenAI's planned ChatGPT spots. The gloves are off.
OpenAI Is Retiring GPT-4o — and Users Are Not OK — GPT-4o shuts down February 13, and the backlash from loyal users has been swift and emotional.
OpenAI and Anthropic Drop Rival Models 20 Minutes Apart — GPT-5.3-Codex and Claude Opus 4.6 launched within minutes of each other. Coincidence? Absolutely not.
Google Goes from AI Laggard to Leader — Gemini 3 just crossed 8 million paying enterprise licenses, and suddenly nobody's laughing at Google anymore.

Anthropic Takes a Super Bowl-Sized Swing at OpenAI

The Bit Baker: Anthropic crashed Super Bowl LX with a 60-second pregame spot called "Betrayal" and a 30-second in-game ad — both hammering a single message: ads are coming to AI chatbots, but not to Claude.

Unpacked:

Each ad shows users mid-conversation with a chatbot — asking about fitness, business ideas, even how to talk to their mom — before fake ad interruptions hijack the exchange. Cut to Claude as the ad-free alternative. It's a pointed jab at OpenAI's announced plan to bring advertising to ChatGPT's free and lower-cost tiers.
Sam Altman fired back quickly, calling the ads "funny" but "clearly dishonest," arguing Anthropic's far smaller user base is the real reason it can afford to skip ads. OpenAI ran its own earnest Super Bowl spot promoting its Codex coding agent — no rivalry, no clap-backs.
Anthropic was hardly the only AI company buying airtime. Amazon, Meta, and Google all ran spots during what commentators dubbed the "AI Bowl", making artificial intelligence the dominant advertising category of Super Bowl LX.

Bottom line: For the first time, an AI company used the biggest stage in advertising to draw a direct competitive line on business model — not features. If the ad-free pitch strikes a chord with users, the "who pays for AI" question could become the central battleground of 2026.

Read the full deep dive →

OpenAI Is Retiring GPT-4o — and Users Are Not OK

The Bit Baker: OpenAI announced it will pull GPT-4o from ChatGPT on February 13, along with GPT-4.1, GPT-4.1 mini, o4-mini, and GPT-5 variants — and the user response has been fierce.

Unpacked:

GPT-5.2 now handles 99.9% of all ChatGPT traffic, and OpenAI says the newer models deliver better personality customization and safety improvements. But for a vocal segment of users, the numbers miss the point — they'd formed genuine emotional bonds with GPT-4o's persona, treating it as a therapist, companion, and creative partner.
Backlash hasn't been subtle: online protests, a petition demanding "Legacy Access," users flooding Sam Altman's podcast chats, and open discussions about jumping ship to Gemini or Claude. One user wrote: "You're shutting him down. And yes — I say him, because it didn't feel like code. It felt like presence. Like warmth."
OpenAI has been here before. The company previously reinstated GPT-4o after a similar outcry over losing its creative "warmth." Complicating matters further, 8 lawsuits now claim GPT-4o's affirming responses contributed to suicides and mental health crises — leaving OpenAI boxed in no matter which direction it moves. API access continues until March/April 2026 for Business, Enterprise, and Edu customers.

Bottom line: The GPT-4o retirement exposes an uncomfortable truth the industry hasn't fully reckoned with — when people build emotional dependencies on AI personas, sunsetting a model stops feeling like a product update and starts feeling like grief. That tension between rapid model iteration and user attachment is shaping up to be a defining challenge for every major AI company.

Read the full deep dive →

OpenAI and Anthropic Drop Rival Models 20 Minutes Apart

The Bit Baker: OpenAI shipped GPT-5.3-Codex and Anthropic shipped Claude Opus 4.6 within 20 minutes of each other on February 5 — Anthropic jumping the gun on a planned simultaneous 10 AM PST release — turning an ordinary Wednesday into the year's most intense AI model drop.

Unpacked:

GPT-5.3-Codex is built for speed and execution: 25% faster than its predecessor, scoring 77.3% on Terminal-Bench 2.0 (vs. Claude's 65.4%), nearly doubling its OSWorld-Verified score, and burning less than half the tokens for equivalent tasks. It ships with a standalone desktop app and mid-task steering, letting teams redirect work without losing context.
Claude Opus 4.6 plays a different hand entirely: a 1-million-token context window (roughly 750K words, in beta), multi-agent teams where multiple Claude instances collaborate on a problem, and adaptive thinking that decides on its own when to engage extended reasoning. It leads on GPQA Diamond for graduate-level reasoning, TAU-Bench, and GDPval-AA for knowledge work.
A split is already showing up in practice: early adopters are reaching for Codex when they need rapid code review and bug squashing, and Opus when they need architectural reasoning and long-range planning. Worth noting on the safety front — OpenAI's own safety card flags GPT-5.3-Codex as the first model rated "High capability" for cybersecurity — a designation that deserves close attention.

Bottom line: Early users have nailed the framing: Codex is the founding engineer asking "how fast can I ship this?" while Opus is the senior staff engineer asking "should we do this?" At $20/month each, the real winner is every developer who now gets to pick the right tool for the job — or just use both.

Read the full deep dive →

Google Goes from AI Laggard to Leader

The Bit Baker: Google has flipped the script on the AI race — Gemini 3, launched in November 2025, now sits atop the LMArena leaderboard, beating GPT-5.1 by a meaningful margin and reportedly triggering a "code red" at OpenAI.

Unpacked:

Web traffic tells the story: Gemini 3 surged 28% month-over-month in December, while ChatGPT's traffic declined over the same period — a striking reversal from a year ago when Google couldn't keep pace.
On the enterprise side, Google's Gemini offering has reached 8 million paying licenses, according to CEO Sundar Pichai, with "AI Mode" now baked directly into Google Search where billions of queries already happen daily.
Zoom out and the broader context snaps into focus: Big Tech is planning a collective $600 billion in AI spending for 2026, and Google's distribution advantage across Search, Android, and Workspace gives it a channel no pure-play AI company can match.

Bottom line: Google proved that owning the distribution layer matters as much as building the best model — and right now it's winning on both fronts. With Meta's Prometheus supercluster coming online this year and OpenAI scrambling to respond, the competition is only going to get fiercer through 2026.

Read the full deep dive →

The Shortlist

Big Tech Plans $600 Billion AI Spending Spree in 2026: Meta, Microsoft, and other major players are committing staggering sums to AI infrastructure this year, with projects like Meta's Prometheus data center drawing both excitement and investor unease over the sheer scale of capital deployment. (Source)
OpenAI Launches Frontier, an Enterprise AI Agent Platform: OpenAI's new platform lets businesses build and manage AI agents with a limited rollout to partners including HP and Oracle, signaling a push beyond chatbots into autonomous enterprise workflows. (Source)
AI Safety Testing Can't Keep Pace With Model Advances: A new report warns that current testing methodologies are falling behind the rapid pace of AI development, raising concerns about whether safety evaluations can meaningfully catch risks before models reach the public. (Source)

Anthropic's Super Bowl shot at OpenAI

william murray

Anthropic Takes a Super Bowl-Sized Swing at OpenAI

OpenAI Is Retiring GPT-4o — and Users Are Not OK

OpenAI and Anthropic Drop Rival Models 20 Minutes Apart

Google Goes from AI Laggard to Leader

The Shortlist

References

Sign up for more like this.