At the TechCrunch Disrupt conference, Amjad Masad, CEO of AI coding platform Replit, dropped major news: Replit has ended its collaboration with Anthropic and fully embraced OpenAI. This model-driven platform once hailed as a star startup now publicly declares “ Claude it’s better now.” For developers writing code with AI daily, this signals a pivotal shift in model selection.ChatGPT It’s better now.” For developers writing code with AI daily, this signals a pivotal shift in model selection.
What happened: Switching from Claude to GPT-6
Masad explicitly stated onstage that the decision to halt collaboration with Anthropic was driven by multiple factors—with cost being critical. Replit announced its partnership with OpenAI in August this year but did not disclose at the time that its relationship with Anthropic had already ended.
His exact words were: “I think ChatGPT it’s better now.” He added that the company uses an internal benchmark to evaluate performance—and also published a benchmark showing OpenAI’s Sol and Astra outperform in “building vibe-coded applications from scratch” with greater stability. “This is a holistic decision based on cost, collaboration, and performance—and right now, they [OpenAI] are the more suitable partner.”
What this means for users
Replit is a cloud-based software development platform where users can build and deploy applications directly in the browser—beloved by founders and developers alike. It previously relied on Anthropic’s Claude model-driven platform, a partnership that significantly boosted the company’s revenue growth.
Now switching to OpenAI models, the most immediate impact is that when Replit users write code or ask AI to generate applications, the underlying “brain” shifts from Claude to the GPT-6 series (Sol/Astra). For average users, the experience change may be subtle; but for developers heavily reliant on specific model characteristics, differences in generation style, code quality, and contextual understanding will be perceptible. To compare the real-world capabilities of these two models, refer to our in-depth review of Claude Sonnet 5.5 and GPT-6.1 Sol .
Controversy and Questions: Conflicting Statements from Both Sides
Masad’s statement is not without controversy. Sources familiar with Replit’s relationship with Anthropic told TechCrunch that the collaboration remains active.
Moreover, evidence suggests Replit was still using Anthropic’s products as recently as weeks ago: a late-September official blog post explicitly mentionedClaude Fable 5 and 5.1 are running in production on the Replit Agent; simultaneously, a Replit product employee was promoting Anthropic’s flagship model Opus 5.5 on X.
TechCrunch has contacted Replit for clarification, but the CEO’s public statements currently stand in clear contradiction with this evidence—leaving “Has Replit fully abandoned Anthropic?” temporarily unresolved.
Industry signal: Platforms shifting from “binding” to “dynamic model selection”
Regardless of the final relationship, this incident reflects a trend in the AI programming space: platforms are moving away from “binding to a single model” toward “switching across multiple models and dynamically selecting based on cost-effectiveness and performance.” With models rapidly iterating, platforms seek an optimal balance of cost, performance, and partnership terms.
For Anthropic, losing a high-frequency use case like Replit is a blow—programming has long been Claude one of its core strengths and among the capabilities most recognized by developers. Conversely, OpenAI’s competitiveness in programming tools is gradually being validated: the GPT-6 series (Sol/Astra) is demonstrating strong performance in real-world scenarios like “vibe coding,” winning platform-level contracts.Claude Anthropic’s accumulated expertise in programming can be seen in Claude Code’s multi-agent projects, while Claude Haiku 5.5 its recent pricing strategy also signals Anthropic’s intensified focus on price competitiveness.
Masad also revealed an interesting detail: Replit’s enterprise business is taking off—enterprise revenue accounted for just 2% last year but jumped to 15% this year, with over half of total revenue now coming from B2B. As enterprise customers become the foundation, model selection logic shifts from “chasing the strongest” to “chasing the most stable, cheapest, and easiest to partner with.”
Frequently Asked Questions (FAQ)
Has Replit fully abandoned Anthropic?
The CEO stated so publicly, but insiders deny it—and just weeks ago, Replit was still using Claude the model in production. Both sides’ accounts currently conflict, and TechCrunch is following up for clarification.
Why is Replit switching models?
The CEO cited a combination of factors: cost, partnership dynamics, and performance. He believes OpenAI’s Sol and Astra deliver superior results in “building vibe-coded applications from scratch.”
Will users notice the change?
For most users, the underlying model switch may not be immediately perceptible; however, developers relying on Claude specific coding styles or capabilities may observe differences in output quality and contextual understanding.
Why did Anthropic spend $11.6 billion on cloud services? Claude Anthropic’s standing in the programming domain?
Short-term impact is limited, but this is a signal—programming platforms are dynamically selecting models rather than locking into a single one. Anthropic must become more competitive on cost and collaboration.
Want to understand how these models truly perform in programming? Check our in-depth review of Claude Opus 5.5,Microsoft-Decision-1 or conduct a side-by-side comparison using Tool Comparison Engine You can also browse Full Review List.