By ByteBot
Replit CEO Amjad Masad walked onto the TechCrunch Disrupt stage on October 16, told the audience that his company was “no longer working with Anthropic,” and called OpenAI’s ChatGPT the better tool. Within 24 hours, he had clarified, reframed, and effectively walked it back. But the business reality underneath the stumble is straightforward: Replit signed a $100M+ annual deal with OpenAI after Anthropic refused to extend a pricing discount, and OpenAI is now the default model powering Replit Agent for millions of developers. The Replit Anthropic OpenAI switch is done — whatever the CEO’s on-stage wording suggested.
What Actually Happened
The story has two layers. The dramatic version — Replit publicly breaks up with Anthropic — is not quite right. Seven Claude models remain in Replit’s catalog, and sources familiar with the partnership told TechCrunch that it was still active even as Masad was speaking. Replit was promoting Claude models as recently as late September 2026.
The accurate version is less dramatic but more consequential: OpenAI offered Replit a discount of more than 30% on a contract worth over $100 million per year. Anthropic refused to match or extend Replit’s existing discount unless Replit committed to higher spending. Replit went with the cheaper option. OpenAI is now the default.
Masad gave the public a cleaner frame after his on-stage comments drew scrutiny. “Replit remains model agnostic,” he told TechCrunch. “Our users can choose Anthropic’s Claude or other supported models. OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership.” The word “economics” is doing a lot of work in that sentence.
The Performance Argument (and Why Developers Are Skeptical)
Masad did not just cite cost. He said that on Replit’s internal benchmarks, OpenAI’s Sol and Astra models “consistently” outperform Claude for building applications from scratch — what Replit’s user base calls vibe-coding. “I think ChatGPT is better now,” he said on stage, a comment that landed awkwardly given Claude Code’s well-documented following among professional developers.
Third-party benchmarks show GPT-6 Astra at 74.1% on DeepSWE v1.1 versus competitive Claude scores. GPT-6.1 Sol, at one-fifth of Astra’s price, reaches 70.8% on the same benchmark. The numbers are not embarrassing for OpenAI. However, Replit’s use case — rapid, prompt-driven app generation for non-expert builders — is not the same workload that Claude tends to dominate in coding-assistant evaluations. The claim that “ChatGPT is better now” depends heavily on which users you’re optimizing for. ByteIota has covered how Anthropic’s API changes have been reshaping developer workflows at the same time this pricing battle was unfolding.
What This Means for Developers on Replit
If you are on Replit’s free plan, you now get OpenAI by default with no option to change it. Free Mode uses automatic model routing, and the model picker is not available at that tier.
If you are on a Core or Pro plan, you can still choose Claude. Open your Agent prompt box, select Power or Max Mode, and pick a Claude variant from the Primary model dropdown. Seven models remain available. According to a detailed breakdown of the model switch, different models consume Replit credits at different rates, so pinning your choice avoids unexpected credit usage if Replit updates its routing logic.
The catch is that most users — paid or otherwise — will never change the default. That is the entire point. When you control the default model on a platform with over 20 million users, you do not need to win every head-to-head benchmark. You just need to be the one that gets called when nobody is paying attention.
The Pricing War Is Real
This deal did not happen in isolation. OpenAI has been cutting prices aggressively throughout 2026 — GPT-5.6 Luna dropped 80% in July, and GPT-6 Sol launched in September at roughly half the cost of its predecessor. Anthropic has countered with its own pricing moves, but the pattern is clear: both labs are willing to operate at steep discounts to lock in enterprise platforms before the market consolidates.
Data from Rippling shows OpenAI’s share of coding work at enterprise companies growing from 16% to 52% over 2026, while Anthropic’s coding share declined from 57% to 18%. Whether that reflects model quality or distribution leverage — defaults on Replit, Cursor, VS Code — is the question nobody can answer cleanly yet. We’ve also seen OpenAI making aggressive moves on multiple fronts this month.
Anthropic’s decision not to match Replit’s pricing will look either disciplined or shortsighted depending on where the AI coding market lands in 18 months. For now, it cost them their seat as the default on one of the most widely used coding platforms on the web. That is a significant piece of infrastructure to concede, whatever the performance benchmarks say.
