Replit Ends Work With Anthropic, Makes OpenAI the Default
Replit is no longer working with Anthropic, co-founder and CEO Amjad Masad said onstage at TechCrunch Disrupt. TechCrunch
Masad cited cost, partnership and performance as the reasons for the decision. Before its deal with OpenAI in August, Replit relied on Anthropic's Claude models to run its platform. Until recently, the company was one of Anthropic's five largest customers.
The switch is complete in production. OpenAI's ChatGPT is now the default model on Replit. OpenAI is now Replit's preferred partner for powering Replit Agent, the tool that handles multi-step coding jobs. Customers can still access Anthropic models on Replit. The company says it remains model agnostic, which means users can choose Claude or other supported models.
The change follows a run of announcements. On August 18, 2026, the company announced Free Mode to expand what is possible with AI, stating that with OpenAI, Replit helps users accomplish more every day. On September 10, 2026, Replit said its Databricks integration is generally available with native Lakebase support. On September 25, 2026, Replit announced it is acquiring Atta to add interactive charts in chat. On September 29, 2026, Replit published work on agent harness design under the title Free the models: Harness design at the frontier, stating Replit Agent lets the model choose its subagents' tier and effort. On October 13, 2026, Replit introduced Replit Solutions Partners to help organizations put code to work. On October 16, 2026, Replit announced Bring Your Own Cloud, offering enterprise isolation and control.
The broader context here is how coding platforms set defaults. Model agnostic does not mean neutral in daily use. Users see a model picker. Below it sits a tuned harness, the support software that manages tools, tests called evals, retries and cost controls. That harness is optimized for one preferred model to run repeated agent loops. Changing it takes real operational work. It affects token use, response time on multi-step tasks, and how work is split when the main model sets the tier and effort for subagents.
On the business side, the cost of running models dominates once agents are used at scale. A top-five customer leaving a model maker gives negotiating power to both sides. For an app company, lower cost per task and closer engineering on the default path can pay for free tiers and longer agent runs. For a model maker, losing a large coding workload removes a rich source of real tool-use records. Coding agents create structured, multi-turn jobs with clear checks, which labs use to improve function calling and planning.
In my view, the enterprise releases deserve as much attention as the model switch. Bring Your Own Cloud separates the control plane from the model plane, so customers keep isolation and control while Replit keeps the agent experience on top. Combined with the Databricks integration with native Lakebase support and a services partner program, the approach recalls earlier platform shifts. Attract developers with a fast default, then add isolation and data connections to carry the same workflow into production. Buyers should test default behavior against selectable models, check data residency under BYOC, and ask how evals track Agent quality as the preferred model changes. If the cost and performance balance holds, this allows more agent time per dollar. That limit has held back agent-based coding tools more than code quality itself, and cheaper, well-tuned runs allow longer plans, more tool calls and more tries before human review.
