Replit Switches Replit Agent Default From Claude to ChatGPT

Replit has switched the default engine behind its coding agent. The Replit Agent Default now points to OpenAI's ChatGPT, demoting Anthropic's Claude to an optional choice in the model selector. The change, announced onstage at TechCrunch Disrupt, is the sharpest signal yet that the agent platform wars will be fought on cost and performance, not loyalty.

The move formalizes a pivot that began in August, when Replit announced a partnership with OpenAI without clarifying what it meant for its long-running relationship with Anthropic. According to TechCrunch's reporting, CEO and co-founder Amjad Masad used the Disrupt stage on Thursday to end the ambiguity, stating that ChatGPT is now the better option for Replit's users.

Performance, cost and partnership decided the Replit Agent Default

Masad attributed the decision to three factors: performance, cost and the strength of the partnership, and those three factors now define how the Replit Agent Default will be judged. According to TechCrunch, he said the company runs an internal benchmark alongside a public one, and that OpenAI's Sol and Astra models consistently outperform on building vibe-coded applications from scratch. The assessment was framed as a practical one: Replit's builders need the model that ships working software fastest, and right now the company believes that is OpenAI's.

The economics of a default setting make the Replit Agent Default change especially consequential for both labs. As TechCrunch noted, most users stick with whatever the platform pre-selects, so shifting the default away from Claude could significantly reduce Replit's spend with Anthropic. The report describes Replit as having been, until recently, one of Anthropic's top-five customers — a relationship that had helped grow the coding platform's revenue substantially, according to reporting by Inc.

Anthropic has not accepted the framing. According to TechCrunch, people familiar with the Anthropic-Replit partnership rejected Masad's onstage comments, with one person saying the partnership remains in place. There is evidence supporting that view on Replit's own properties: a blog post from late September cited Anthropic's Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent's production environment, and a product employee had posted on X about using the flagship Opus 5.5 model around the same time. After the Disrupt appearance, Masad clarified that customers can still access Anthropic models on the platform — and Replit's October 16 changelog lets Core and Pro builders select Claude Haiku 5.5 in Power Mode.

Masad later told TechCrunch over the weekend that Replit remains model-agnostic, describing OpenAI as the preferred partner for powering Replit Agent based on performance, economics and the strength of the partnership. According to WebProNews's account of the weekend clarification, the CEO framed the move as a default-setting decision rather than an outright breakup.

A "self-driving company" and the habitat agents need

Beyond the model swap, Masad used the Disrupt stage to sketch where Replit is heading as a business. According to TechCrunch, he described the company as a "self-driving company," with agents handling much of the routine work while people supply judgment and taste. Enterprise is now about 15 percent of revenue, up from 2 percent a year ago, and more than half of revenue is B2B. Backstage, he told TechCrunch he expects more than half of revenue to come from enterprise and government customers in 2027.

That enterprise pivot raises the stakes of the Replit Agent Default decision. The models that power coding agents inside government and corporate environments will be chosen for compliance, auditability and cost at scale — the same procurement logic that has already reshaped how companies pick the infrastructure their agents run on, from GitHub Copilot's model rotation schedule to the wave of governed agent frameworks hitting regulated industries.

Masad also offered a framing that will likely outlive this particular partnership shuffle. According to TechCrunch, he said that just as humans need an environment to work in — offices, food, air conditioning, computers — agents need a habitat too. The habitat he described has good orchestration, good error recovery, good safety and security, good sandboxing, good networking, and good identity and authorization. Once that habitat exists, he argued, agents can be deployed safely. The metaphor lands because the industry keeps learning it the hard way: as websites increasingly close their doors to AI agents, the question of where agents are allowed to operate — and under whose identity — has become as important as what they can do.

What the switch signals for the agent platform wars

Three things stand out in the Replit Agent Default story. First, model defaults are now commercial weapons: the difference between "default" and "available in the selector" is the difference between paying Anthropic and paying OpenAI at Replit's scale. Second, the partnership era is over and the performance era has begun — platforms that once marketed co-development deals are now benchmarking publicly and switching on the numbers. According to AIPulse's news roundup, Replit's weekend clarification positioned the change as driven by cost considerations, performance benchmarks and a new strategic partnership, not ideology.

Third, the Anthropic insiders' pushback is a reminder that in the agent economy, a partnership can be simultaneously over and alive. Watching the Replit Agent Default over the next quarter will show whether this was a pricing renegotiation or a permanent realignment. Replit still lists Anthropic models, still lets users pick Claude, and still describes itself as model-agnostic — while routing its largest single stream of inference spend to OpenAI. The Replit Agent Default is less a divorce than a renegotiation conducted in public, with the default setting doing the talking.

For builders, the practical takeaway is the oldest one in platform economics: watch the default, because the Replit Agent Default is where the money moved. The model your users reach when they do nothing is the model that wins — and Replit just changed what "nothing" means.