A heated rivalry

TL;DR: OpenAI and Anthropic spent the week trading blows, from dueling product launches and talent poaching to a very public fight over a Super Bowl ad. But beneath the spectacle is a divide that goes beyond battling for market share: The two companies hold very different visions for how powerful AI models should be built and governed.

What happened: From the industry that brought us Apple vs. Google, Edison vs. Tesla (the other one), and that weird Musk vs. Zuck MMA cage fight that never actually happened, comes a new rivalry: Anthropic vs. OpenAI. While the competition between the two AI companies runs deep, this week they served up a dizzying game of one-upmanship that felt like watching the final ping-pong—sorry, table tennis—match in Marty Supreme. (Or, perhaps, a tense standoff in the ice hockey rink.)

Here’s a play-by-play: On Monday, OpenAI made the first volley by dropping a macOS app for Codex, which is OpenAI’s coding agent. It felt like a direct shot at Claude Code, the agentic coding MVP that’s taken the software world by storm.

On Tuesday, Sam Altman revealed that OpenAI had poached Dylan Scandinaro, a former Anthropic safety researcher, to be its new “Head of Preparedness” (we’re still not sure what that means). It was a reversal from the usual flow of OpenAI employees jumping ship to Anthropic.

Wednesday, however, was OpenAI’s Waterloo. Anthropic unveiled a series of Super Bowl ads running this weekend—all of them mocking OpenAI’s recent 180 on putting ads in ChatGPT. Altman’s response? Totally chill. He only posted a 421-word diatribe (we miss the old Twitter character limit) on X accusing Anthropic of using Orwellian “doublespeak” and called it an “authoritarian company” for blocking other companies from using its AI model. Altman also bragged that people were switching to Codex (presumably from Claude Code, but only Altman knows), claiming that there had been 500,000 downloads since Monday and that he believes “Codex is going to win.” Altman said Anthropic’s ads made him laugh, but we’re going to need a Snopes fact check on that.

On Thursday, there was a one-two punch from OpenAI. First, it launched Frontier, an enterprise platform for building, deploying, and managing AI agents in a bid to lock in more corporate customers. A few hours later, Anthropic dropped Opus 4.6, an update to its AI model that focuses on deeper reasoning and enterprise workflows. Less than an hour after that, OpenAI landed the other blow: an update to its own agentic coding model, GPT-5.3-Codex, that claims to be 25% faster than 5.2. OpenAI emphasized that Codex’s latest model had basically coded itself—something that Anthropic bragged about with Cowork weeks ago.

The feud so far: Just think of Anthropic vs. OpenAI like the Western Schism. Anthropic came about as an offshoot of disillusioned OpenAI employees; Dario Amodei served as VP of research at OpenAI before departing in late 2020 with his sister Daniela (VP of safety and policy) and roughly 14 senior researchers to found Anthropic. Now there are two AI popes—or at least two competing doctrines for how a frontier model company should be run. For Amodei, the rift came down to a fundamental disagreement: OpenAI’s model capabilities were accelerating faster than its safety work, and Microsoft’s multibillion-dollar investment introduced commercial pressures that made that imbalance harder to correct.

OpenAI tends to ship model updates first, then tighten guardrails later (though Anthropic has also moved at a fairly whirlwind pace lately). And while both companies have safety boards, the OpenAI board’s failure to oust Altman in 2023 has left some doubt around its power to veto key decisions. Meanwhile, Anthropic has a reputation as the more philosophical, mission-driven “good AI” shop—an image it has played up, even revising Claude’s code of ethics last month, though there are growing concerns of burgeoning contradictions in its mission.

And while Anthropic is the underdog—ChatGPT now has more users on mobile than Claude, Gemini, Grok, and two Chinese models put together—that hasn’t stopped it from punching above its weight lately. Claude Code was an electric shock felt throughout the software world, inspiring looser, more permissive (and notably less secure) clones like OpenClaw—and ushered in this era of AI agents who can code and run your entire life, which led (in part) to a tumble in software stocks this week.

The real winner of this beef: Google, whose AI-integrated kingdom is bringing in record revenue, can sit comfortably on the sidelines and watch its competitors punch each other. Oh, and also the short sellers who pocketed $24 billion betting that software stocks would fall. —WK