AI Agent UpdateReplit ditches Anthropic for OpenAI as its default coding agent
One of Anthropic's biggest customers just changed its default. Replit CEO Amjad Masad used a TechCrunch Disrupt stage to announce that ChatGPT now powers Replit Agent — a cost-and-benchmarks decision that says plenty about where the model wars are actually fought.
The News
Replit CEO Amjad Masad said onstage at TechCrunch Disrupt on Oct 17 that OpenAI's ChatGPT is now the default for Replit Agent, displacing Anthropic in the slot that decides what the overwhelming majority of users actually run.
Masad's stated reasons were cost, partnership, and performance — specifically on Replit's public vibe-coding benchmark, where OpenAI's Sol and Astra models beat Claude. This wasn't a quiet config change buried in a changelog; it was announced from a conference stage, with the reasoning attached.
Anthropic's models remain selectable inside Replit — users who want Claude can still choose it. But defaults are the whole game in platform economics: the preselected option captures nearly all the usage, and with it nearly all the inference revenue.
The size of the account is what elevates this from product news to industry signal. Replit was reportedly a top-five Anthropic customer by usage — the kind of flagship deployment the labs put on their own keynote slides.
Default-slot economics are the new battleground — OpenAI just took a flagship account from Anthropic on cost and benchmarks, not vibes.
Why It Matters
In the API era, model choice was a developer decision made repo by repo. In the agent era, it's a platform decision made once — by whoever owns the default. That concentrates enormous inference volume into a handful of negotiable slots, and turns every major agent platform into an auction house for model providers.
For Anthropic, losing a top-five customer's default is a revenue and narrative hit at an awkward moment — the same week it is circulating a draft federal safety bill and watching the agent-safety debate move without it. For OpenAI, the win validates a cost-plus-benchmarks pitch it can now carry to every other platform currently defaulting to a rival. And because Replit's users are exactly the developers every lab wants to habituate early, the default there carries an outsized signalling weight for the whole category.
Analyst's Take
Default-slot economics are the new battleground. The agent era turns model choice into a platform decision, and OpenAI just proved it can take a flagship account from Anthropic on cost and benchmark performance — not just vibes, not just brand. That is a much more dangerous kind of competition for the incumbent, because it's repeatable, and it scales with every platform that re-runs the same benchmark.
Watch whether other agent platforms start auctioning their defaults — and what the winning bids look like. If defaults become a paid or discounted placement market, model gross margins compress the way search-default economics compressed them for everyone except the platform owner. The quiet winner in that world isn't a lab at all: it's whoever owns the surface where the default gets set.
By The Numbers
FAQ
What did Replit announce at TechCrunch Disrupt?
CEO Amjad Masad announced that OpenAI's ChatGPT is now the default model for Replit Agent, replacing Anthropic. He cited cost, the partnership, and OpenAI's Sol/Astra models beating Claude on Replit's public vibe-coding benchmark.
Can Replit users still use Claude?
Yes — Anthropic's models remain selectable in Replit. What changed is the default, which is where the overwhelming majority of usage, and therefore inference spend, concentrates.
Why does a default model slot matter so much?
In agent platforms, the preselected model captures nearly all usage. Winning the default at a major platform is worth more than winning benchmark headlines, because it converts directly into sustained inference volume and revenue.
How big a customer was Replit for Anthropic?
Reportedly a top-five Anthropic customer by usage — which is why the switch is being read as the model wars' clearest platform-level defection yet.
Source: TechCrunch