Replit, one of Anthropic's top five customers, makes OpenAI the default for Agent: cost, terms and 'I think ChatGPT is better now'
Amjad Masad said it onstage at TechCrunch Disrupt on October 16: Replit is no longer working with Anthropic. The reasons given were cost, partnership terms and performance, with the line that stuck being I think ChatGPT is better now. Replit's internal benchmark and its public one both show Sol and Astra doing better at building vibe-coded apps from scratch, according to Masad, who added that Replit had until recently been one of Anthropic's top five customers. TechCrunch ran it at 21:22 UTC under the headline Replit is no longer working with Anthropic, then spent the next day walking it back with sources.
The walk-back came from both sides. People familiar with the partnership told TechCrunch the Anthropic deal is still in place. Replit's own blog named Claude Fable 5 and 5.1 as production models in Agent as recently as late September, and a Replit product employee was promoting Opus 5.5 around the same time. At 16:29 UTC on October 17 Masad posted a clarification: Replit remains model agnostic, users can still pick Anthropic models, and what was meant is that OpenAI is now the preferred partner powering Agent based on performance, cost and strength of the partnership, optimizing the default experience for the best performance at the best price. TechCrunch retitled the piece to Replit makes OpenAI its default over Anthropic at 18:48 UTC. Replit announced an OpenAI partnership in August, so the switch was months in the making.
Default is the operative word. Most customers never change it, so moving the default moves the spend, and a top-five customer moving its default is a revenue event for Anthropic whether or not the contract technically survives. It is also the cleanest public datapoint yet on model churn inside agent products: Replit's own September post argued the harness does not disappear when models improve, and here the harness stays while the model underneath it swaps out on price and terms. The Disrupt numbers around it: enterprise went from about 2 percent of Replit revenue last year to about 15 percent this year, more than half of revenue is now B2B, and Masad expects over half from enterprise and government in 2027. Masad also laid out what agents need before safe deployment, calling it a habitat: orchestration, error recovery, sandboxing, networking, identity and authorization controls. That list is the agent-infrastructure market in one sentence, and it is a list of things a model vendor does not sell.
Anthropic had not commented by Saturday night. The timing is awkward for it on two other fronts: the Anthropic Institute's employment report, published the same afternoon, says agents inside Anthropic now run six hours for every human hour, and the FT reported the same day that KBRA rated Anthropic BB+ in June, one notch below investment grade, ahead of its $35 billion chip financing. A top customer saying the competitor is cheaper and better on its benchmark is the kind of sentence a rating analyst reads twice.
Links: https://techcrunch.com/2026/10/17/replit-is-no-longer-working-with-anthropic-ceo-says/ , https://x.com/amasad/status/2111494912562888939 , https://techcrunch.com/2026/10/16/replits-ceo-says-the-future-of-work-is-the-self-driving-company/
← Back to all articles
The walk-back came from both sides. People familiar with the partnership told TechCrunch the Anthropic deal is still in place. Replit's own blog named Claude Fable 5 and 5.1 as production models in Agent as recently as late September, and a Replit product employee was promoting Opus 5.5 around the same time. At 16:29 UTC on October 17 Masad posted a clarification: Replit remains model agnostic, users can still pick Anthropic models, and what was meant is that OpenAI is now the preferred partner powering Agent based on performance, cost and strength of the partnership, optimizing the default experience for the best performance at the best price. TechCrunch retitled the piece to Replit makes OpenAI its default over Anthropic at 18:48 UTC. Replit announced an OpenAI partnership in August, so the switch was months in the making.
Default is the operative word. Most customers never change it, so moving the default moves the spend, and a top-five customer moving its default is a revenue event for Anthropic whether or not the contract technically survives. It is also the cleanest public datapoint yet on model churn inside agent products: Replit's own September post argued the harness does not disappear when models improve, and here the harness stays while the model underneath it swaps out on price and terms. The Disrupt numbers around it: enterprise went from about 2 percent of Replit revenue last year to about 15 percent this year, more than half of revenue is now B2B, and Masad expects over half from enterprise and government in 2027. Masad also laid out what agents need before safe deployment, calling it a habitat: orchestration, error recovery, sandboxing, networking, identity and authorization controls. That list is the agent-infrastructure market in one sentence, and it is a list of things a model vendor does not sell.
Anthropic had not commented by Saturday night. The timing is awkward for it on two other fronts: the Anthropic Institute's employment report, published the same afternoon, says agents inside Anthropic now run six hours for every human hour, and the FT reported the same day that KBRA rated Anthropic BB+ in June, one notch below investment grade, ahead of its $35 billion chip financing. A top customer saying the competitor is cheaper and better on its benchmark is the kind of sentence a rating analyst reads twice.
Links: https://techcrunch.com/2026/10/17/replit-is-no-longer-working-with-anthropic-ceo-says/ , https://x.com/amasad/status/2111494912562888939 , https://techcrunch.com/2026/10/16/replits-ceo-says-the-future-of-work-is-the-self-driving-company/
Comments