Situation report: a public breakup claim from one of AI coding’s biggest platforms, and the other side disputes it.
Replit CEO and co-founder Amjad Masad said Thursday onstage at TechCrunch Disrupt that his company is no longer working with Anthropic. According to TechCrunch AI, Masad pointed to cost, partnership and performance, and said OpenAI’s models now do the job better. Within hours, people familiar with the deal pushed back and said the partnership is still in place.
That contradiction is the story. Here’s what we know.
🎯 The Claim
- Who: Amjad Masad, CEO of Replit, the browser-based platform where users build and deploy apps, often through vibe coding (describing what you want in plain language and letting an AI agent write the code).
- What: Replit has stopped working with Anthropic, according to Masad.
- Why, per Masad: “I think ChatGPT is better now.” He said Replit’s internal and public benchmarks show that “Sol and Astra perform better on building vibe-coded applications from scratch.”
- His summary: “It was a decision based on both cost, partnership, and performance. And for all these reasons, we just felt like they were a better partner for now.”
- Background: Replit announced a partnership with OpenAI in August. Nothing at the time suggested it was walking away from Anthropic.
⚠️ The Counter-Evidence
TechCrunch AI reports that insiders familiar with the partnership rejected Masad’s account. One source, who asked not to be named, said the deal is still active. The public record also complicates his story:
- A Replit blog post from late September lists Anthropic’s Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent’s production environment.
- Around the same time, a Replit product employee posted on X praising Anthropic’s flagship model, Opus 5.5.
- TechCrunch has asked Replit to clarify and hasn’t heard back yet.
So you can read this two ways. Replit may have moved its default workloads to OpenAI without formally ending the contract. Or a CEO said more onstage than the contracts actually support. Either way, Replit’s production stack has clearly changed a lot in just a few weeks.
📊 Why It Matters
This matters because Claude models helped Replit grow. TechCrunch AI notes that the Anthropic partnership was said to have driven significant revenue growth. When a flagship customer publicly says a competitor is “better now,” that’s a signal to the market, whether or not the paperwork has changed.
The bigger takeaway is that model loyalty is thin. AI coding platforms now treat foundation models as interchangeable parts. They benchmark them constantly and switch whenever the cost or performance numbers move. Model providers are competing on price per token and on coding benchmarks at the same time, and their customers will happily go public about it.
💼 Replit’s Business Shift
Masad also gave a rare look at where Replit’s revenue comes from:
- Enterprise was about 2% of revenue last year.
- This year, it’s about 15% from enterprise contracts.
- More than 50% of revenue is now B2B.
- He’s still not sharing ARR figures.
That shift helps explain the vendor drama. Enterprise customers push hard on cost and reliability, and model spend is a big line item for any AI coding platform.
🛡️ Masad on Agent Safety
Masad also argued that agents need a proper “habitat” before anyone can deploy them safely. In his view, that means “good orchestration, good error recovery, good safety and security and sandboxing, good networking, good identity and authorization.” He called it the agent’s version of an office with air-conditioning and decent food.
🧭 What to Watch
- A clarification from Replit. Watch whether it confirms a full break or walks the comment back to a shift in its default model.
- A response from Anthropic. So far, the denial has come only from anonymous insiders.
- Changes to Replit Agent’s model list. If Claude disappears from production, that settles it.
- Other coding platforms making similar moves. Expect more public benchmark comparisons used as leverage in negotiations.
If you build on Replit, keep an eye on output quality and pricing over the next few weeks. A model swap underneath your agent can change results without warning. TechCrunch AI has the full report and says it will update the story as Replit responds.