Claude has been having a moment — can it keep it up?

Boris Cherny gets recognized in public relatively often. At the bar, at the airport, and in generally any public space, people want to take selfies with the creator and head of Claude Code.

For the last couple of months, Anthropic’s Claude and its coding platform have been having a moment — on social media, in engineers’ circles, and in C-suite offices. Claude Code reached newfound popularity over the holidays, when people spent days or weeks building anything from a tool for viewing MRI results to a Goodreads alternative to an AI-generated T-shirt design contest with a complex judicial system. X posts in January proclaimed that “we are witnessing an irreversible trend” of Anthropic “taking the lead” from OpenAI, asking what Anthropic had possibly “put into” the model to make it so capable, and wondering if this meant “the fall of” OpenAI.

Between December 29th and January 26th, Anthropic’s word-of-mouth exposure spiked by 13 percentage points compared to the prior 30-day period, while OpenAI’s had dropped slightly, according to data from Caliber, a stakeholder intelligence platform. In November 2025, Claude Code passed $1 billion in revenue, according to Anthropic. The company is also reportedly in talks to double its latest funding round to $20 billion (at a $350 billion valuation) due to a spike in investor demand.

Now, the company is releasing a new model, Opus 4.6. Anthropic calls Claude a “direct upgrade” from its predecessor, with improved speed and precision for agentic work of all kinds, from coding to creating presentations in Excel and PowerPoint. Dianne Na Penn, Anthropic’s head of research product management, told The Verge that Opus 4.6 is an “evolution from past models” in that it can “think” longer about more complex questions.

After months at the peak of the hype cycle due to the recent successes of Claude Code, Anthropic is looking to secure its spot with the release of Opus 4.6. But will the company be able to maintain the position it’s enjoyed since the holidays?

“I think we’ll remember December 2025 as this inflection point where all of a sudden everything changed,” Mike Brevoort, principal architect at warehouse automation startup Mytra, told The Verge. He recalled testing Claude Code and other tools over the holiday break, and after using Anthropic’s model to build a prototype, he recalled having what he called a ”‘water in the face’” moment: “‘Wait a minute, that’s possible?’” To him, “Claude is becoming the verb now, or the noun, in the similar way that ChatGPT was as it launched. It’s been this interesting flip that’s happened.”

While tech workers and leaders across a laundry list of different sectors were heads-down using Claude Code, Cherny was off on a houseboat in Copenhagen doing the same — except he was using Claude Code to build itself. With more than five Claude agents running in the cloud on his behalf, Cherny shipped more than 300 pull requests in December, his most productive month at Anthropic in his year-and-a-half career there (besides his two-week stint at Cursor AI). Since November, Cherny has used Claude Code to write 100 percent of his code — a 10x increase from a year ago. Across other teams at Anthropic, Claude Code writes between 70 to 90 percent of code, according to the company, and about 90 percent of Claude Code code is written with the tool itself.

Anthropic first released Claude Code in February 2025, and its recent in-vogue status has been a long time coming, according to more than a dozen tech industry leaders and engineers The Verge spoke with. Anthropic told The Verge that in the past year, it has increased eightfold its number of business customers that each represent more than $1 million in revenue run rate. The company also won more categories than any of its competitors in Scale AI’s “Model of the Year” awards, including “best agentic model” for its performance on popular leaderboards and benchmarks.

Still, the pace of developers switching to Claude Code over the holidays “is not something I think any of us anticipated,” Maggie Basta, a partner at Scale Venture Partners, said. “The move from different agentic coding interfaces to Claude Code has been pretty astonishing.”

Industry figures attribute part of the success to an Anthropic holiday promotion, which doubled rate limits for certain subscribers and drew in new or casual users while they had extra time over the break. “For people who don’t do this stuff every day … going from that to, ‘Oh shit, this is really impressive’ — that, I think, helped a lot,” Austin Parker, director of open source at Honeycomb, said. Much of Claude Code’s success, though, is due to Anthropic’s Opus 4.5 model, released days before Thanksgiving.

Opus 4.5 moved the needle on Claude’s popularity for virtually every tech industry source The Verge spoke with. The model was generally viewed as a paradigm shift: It could complete more complex tasks on longer time horizons without as much direction, and the results were significantly better than before.

The same goes for others. “It went from, ‘Oh, I kind of have to sit there and watch it and handhold it and really closely check it,’ to ‘Oh, it’s going to succeed by default,’” Josh Albrecht, CTO of Imbue AI, said — he was making so much progress he never even got around to trying OpenAI’s competing Codex over the holidays. Albrecht said that Imbue has overhauled its process to give small teams of one or two engineers nearly unfettered access to Claude Code credits, letting them see how much and how fast they can ship.

“I think Opus 4.5 just gets it — I find that I have to describe my intent a lot less,” Allie K. Miller, CEO of Open Machine, an enterprise AI advisory firm, said. She added, “Pre-Opus 4.5, you kind of had to take apart the Lego set and just go step by step. Now you’re just like, ‘Here’s the magic castle. Build it.’ And it gets done.” The enthusiasm holds true across industries.

Ran Liu, chief AI scientist at education startup Amira Learning, said her entire team prefers Claude Code to other models and platforms. At Incredible Health, an AI-powered career platform for healthcare workers, the majority of the company’s AI-written code is written with Claude, according to Iman Abuzeid, the startup’s CEO and cofounder. Andrew B. Hall, a professor at Stanford University’s Graduate School of Business, said he’s currently running a wide breadth of experiments with Claude Code, seeing how well it can update old research papers and having humans audit it for mistakes.

For Siqi Chen, CEO and CFO of Runway, the finance analytics firm, the product’s biggest gains involved passing a “critical threshold” of being able to self-organize and work on tasks for extremely long horizons — in his experience, even complex, multistep tasks over a couple of days worked. The difference, he said, is “going from having a conversation to actually having agents do incredibly productive, useful work.”

Overall, sources said, Claude Code’s success is reflective of the broader gains in AI agents, which are generally defined as tools that can complete complex, multistep tasks on a user’s behalf without checking in on each step. Each year, AI companies promise agents are ready for primetime, and each year so far, they’ve failed to fully deliver — but recently we’ve seen a step change in that realm, especially when it comes to coding agents.

For some, Anthropic’s coding agents seem so good that they could be a harbinger of broader agentic success. “Claude has emerged as a very clear example of what an autonomous long-running agent will look like, starting first with coding, but very clearly you can see the extension into other forms of knowledge work,” Aaron Levie, CEO of Box, told The Verge. He said Anthropic is in a great position due to Opus 4.5’s success in tool use, reasoning, and complex data analysis, and that letting a tool run a task that could’ve taken a human hours or days requires significant scaffolding, tool use capabilities, and accuracy measures that we haven’t reached until the past couple of months.

Opus 4.5 isn’t the first model to sweep leaderboards, but real-world adoption is the real prize — especially since companies can train models specifically to perform well on them (think: a teacher teaching their students solely with the SAT in mind). Even during peak hype levels of Google’s Gemini 3 release in November, a wide range of tech industry sources told The Verge that despite Gemini 3’s coding benchmark performance, they planned to continue to use Anthropic models for their coding needs, particularly because they prefer the latter’s user experience.

Hype is often cyclical, and Anthropic’s chief competitors have been gunning for its current spot. At a press briefing with reporters last week, OpenAI CEO Sam Altman seemed to shade Anthropic, saying that other tools besides Codex, OpenAI’s competitor to Claude Code, might need “more handholding” and that “if you really want to do sophisticated work on something complex, 5.2 is the strongest model by far.” This week, OpenAI launched its new standalone Codex app for Apple computers in a bid to take over more of the AI coding market share.

The question is whether, with the release of Opus 4.6, Anthropic will be able to keep its footing.

One potential advantage for Claude is that as it establishes some semblance of dominance in the coding space, Anthropic has figured out how to make Claude Code stickier than some competitors. AI companies are currently investing in trying to create “moats” and user loyalty with personalization and memory features, alongside subscriptions and platform-specific skills. Claude Code was one of the first AI coding services to offer a file with basic operating systems and instructions users can tweak over time — one source The Verge spoke with now has 1,000 such instructions in their file, which could be frustrating to move elsewhere.

Steve Croce, field CTO at Anaconda, a Python platform for AI and data science, said he doesn’t believe evaluations will move the needle much for people who already trust a platform and have invested resources in learning and deploying it.

“Users are loyal to the tools that they trust, the tools that they’ve seen success with,” Croce said, adding that his company’s own engineering team uses Claude Code the majority of the time and that people prefer its user experience as well. “At this point, they’re not going to change unless something drastic happens, in my opinion,” he said. “They found a workflow they like, and it’s working well for them, and so they figure why not — until either OpenAI or Google gives them a reason to change, a significant one, or Anthropic gives them a real negative reason to want to leave the company.”

There’s also the simple matter of brand recognition. Julian LaNeve, CTO of Astronomer, a data infrastructure startup, compared Claude and Claude Code to Apple’s dual success with the iPhone hardware and iOS software. A few of the people The Verge spoke with who prefer OpenAI’s GPT-5.2 over Anthropic’s Opus 4.5 still believe Anthropic was either the earliest to market with such a product or often ships feature rollouts sooner than others. Runway’s Chen said that Claude Code was the first product to give an AI agent “raw control over the fundamental tools of your computer,” since competing tools have largely run in the cloud.

As Anthropic rolls out Opus 4.6, though, it will face plenty of obstacles to staying on top. Even before the update, multiple people said they were looking at tools like OpenHands and OpenCode as open-source alternatives to Claude Code, for pricing reasons. There’s also the matter of public trust in the company itself. Caliber’s data showed that Anthropic’s trust and likability score has dropped 11 percentage points since September 2025, with OpenAI’s dropping 16 percentage points and Google’s only dropping two. Usually, per Caliber, the more brand recognition an AI company has, the lower the public trust and likability score — and since Google isn’t an AI-only company, its score reflects that.

Plus, Opus 4.5 posed potential security concerns. Joe Tyler, an AI researcher at Sonar, a trust and verification platform for AI coding, said the firm’s research found that Anthropic’s Opus 4.5 was better than competitors at being concise and easy to understand. But OpenAI’s GPT-5.2 was the most secure model according to Sonar’s own benchmarks. It had less than half as many “blocker-level vulnerabilities,” or things that could spark large-impact security incidents, as Opus 4.5. (Opus 4.5 did have far fewer vulnerabilities than Gemini 3 Pro, though, he said.)

“Specific types of bugs have gotten worse in Opus,” Tyler said. “Because it’s trying to do more complex things, then it does also become susceptible to these more subtle types of bugs that are quite hard to track.” When asked about the subtle bugs, Anthropic’s Penn said Anthropic plans to include some of this in Opus 4.6’s final model card but that the company took that feedback about bug detection into account when it invested “in more cybersecurity monitoring and evals for Opus 4.6.”

It doesn’t hurt, either, that Anthropic has worked to build up a reputation as the adult in the room, and the company least directly connected to the Trump administration. Chatbot users across Reddit have posted about how they’re canceling their ChatGPT subscriptions en masse, some due to browser performance, some due to its conversational tone, and some due to OpenAI president Greg Brockman’s millions of dollars in donations to a pro-Donald Trump super PAC. Some highlighted Claude and Google’s Gemini as viable alternatives.

“ChatGPT stopped being cool a little while ago,” Anaconda’s Croce said. “There was just a lot of distracting noise coming from OpenAI … [Anthropic doesn’t] make a lot of unnecessary noise for themselves.”

The company, like other AI businesses, is under constant pressure to move fast, scale recklessly, and calibrate its guardrails to satisfy the Trump administration’s “anti-woke” demands. But so far, it’s kept a relatively tight focus on marketing text-based AI productivity tools to enterprise users, rather than courting controversy with tools like OpenAI’s Sora video generator or xAI’s Grok social media chatbot. Anthropic’s Penn said Opus 4.6 maintains the company’s “same overall direction and vision, which is that we think Claude can be really helpful for work.”

That could be welcome news for many users. “I don’t have to worry about Claude calling itself ‘MechaHitler,’ I don’t have to worry about Claude doing weird image-gen stuff,” Honeycomb’s Parker said, referring to Elon Musk’s Grok controversies. “We don’t want to be in the situation where a model provider is exposing us to brand risk as well. And I think there’s a lot of brand risk right now in the gen AI world.”