Replit Walks Away From Anthropic, Points to OpenAI on Cost and Performance
Replit CEO and co-founder Amjad Masad said onstage at TechCrunch Disrupt 2026 that Replit is no longer working with Anthropic. The remark was reported on Oct. 16, 2026, the same day Masad appeared on the Disrupt Stage TechCrunch.
Masad attributed the break to three factors: cost, partnership and performance. Discussing Replit's current AI partners, he added, "I think ChatGPT is better now." Replit announced a partnership with OpenAI in August 2026.
The onstage claim conflicts with another account in the same report. A person familiar with the Anthropic-Replit partnership, speaking anonymously, told TechCrunch the partnership was still in place. Neither account includes additional detail on contract status, API usage, or deployment timelines.
Until recently, Claude was central to Replit's stack. Under its partnership with Anthropic, Replit used Claude models to run its platform. A September 2026 Replit blog post cited Anthropic's Claude Fable 5 and Claude Fable 5.1 as models running in Replit Agent's production environment.
That production dependency has deep roots. Replit integrated Claude into Agent to turn natural language into code, an integration Anthropic later said drove 10X revenue growth for Replit. Replit used Claude 3.5 Sonnet's computer use and UI navigation capabilities to develop a feature that evaluates apps. In its Claude 4 announcement, Anthropic said Replit reported improved precision and dramatic advancements for complex changes across multiple files.
Replit's infrastructure footprint extends beyond a single lab. Replit announced a partnership bringing vibe coding to Microsoft Azure, and was among the first to evaluate Anthropic on Azure Databricks via the Mosaic AI gateway. Replit uses Claude on Google Cloud's Vertex AI and on Google Cloud's Agent Platform to build and deploy software applications. Its AI adoption writing cites Anthropic's Building effective agents as a practical vocabulary for agent architecture including tools, retrieval, memory and evaluation.
The model switch coincides with a shift in Replit's customer mix. Masad said enterprise was 2% of Replit's revenue last year and 15% this year is enterprise contracts. He said more than 50% of Replit's revenue is B2B. Replit says it has more than 50 million creators.
Masad's planned Disrupt discussion topic was his perspective on the future of programming and Replit's role in developing it. In a separate Disrupt session, he described that future as "self-driving companies" where AI agents handle day-to-day tasks TechCrunch. Masad has previously described Replit's advantage over Anthropic as its nimbleness and focus as a smaller company.
Looking at what this means for teams operating agent-based coding platforms, the dispute is less about a single vendor and more about procurement leverage. For an expert operator, model selection is now a routing problem across price per token, inference latency, context handling, tool-use reliability, and eval scores on multi-file edits. A platform with a growing enterprise book can negotiate harder on all five, and can justify re-tuning prompts, harnesses and guardrails if another lab offers better unit economics or steadier throughput.
In this author's view, the episode points toward shorter model allegiances. Worth flagging is how quickly production references can age. A September reference to Fable 5 and 5.1 in Agent does not settle October deployment, just as an August partnership announcement does not by itself explain October performance claims. The durable change is structural. When natural-language-to-code becomes the interface, the provider that controls distribution to developers and to enterprise buyers gains room to swap the underlying engine without asking users to change behavior. That is what makes multi-model operation possible at scale, and it is likely to keep pricing and partnership terms in flux.
