Replit makes OpenAI default, keeps Claude as option

By VK

Published 18 Oct 2026, 09:46 UTC

Reporting from TechCrunch

Replit CEO Amjad Masad said it had dropped Anthropic, then clarified OpenAI is now the default and Claude models stay selectable.

What it means for founders

Pin the model you trust. If you build with Replit Agent and liked what Claude produced, open the selector in Power or Max mode and choose it explicitly. A quiet change of default can alter code style, quality and credit burn from one sprint to the next.
Defaults are the real platform risk. Replit can redirect a large share of its spend from one lab to another by changing a single setting. If your product wraps a model, treat the default provider as a choice that pricing, partnership terms or a benchmark result can reverse fast, and keep a second provider tested.
Price agents per task, not per token. Replit says Max mode scores 72% on the DeepSWE benchmark at about $2.11 per task. Run your own evals on cost and success rate before locking an agent product to one model.
Watch for formal statements. The signals to track are any official word from Anthropic or Replit on the agreement, whether the Claude list in Replit's selector shrinks, and whether Replit publishes revenue figures as enterprise grows toward its 2027 goal.

The story

Replit chief executive Amjad Masad told a TechCrunch Disrupt audience on Thursday, October 15, that his AI coding company had stopped working with Anthropic, pointing to cost, its OpenAI relationship and model quality. After the session he narrowed that claim: OpenAI models are now the default inside Replit Agent, while Anthropic's Claude models stay available to anyone who selects them. TechCrunch later revised its report and headline to reflect the clarification.

Why Replit moved its default to OpenAI

Masad argued that OpenAI's current models beat Anthropic's at the task Replit cares about most, turning a blank prompt into a working app. He cited an internal benchmark plus a public one, naming OpenAI's Sol and Astra as the steady leaders. In August Replit launched Free Mode, which runs on OpenAI's GPT-5.6 Luna and lets subscribers on the $20 Core plan handle routine work without spending credits. A late September engineering post, Free the models, describes GPT-6 Astra as the core loop behind Max mode and compares it with Claude Fable 5 and Fable 5.1 in production traffic.

"Replit remains model agnostic," Masad later told TechCrunch. The company's model selector documentation still lists Claude Sonnet 5, Fable 5.1 and several Opus versions in the paid Power and Max modes, next to GPT-6 Sol, GPT-6 Astra and Kimi K3.

What the switch means for Anthropic

Defaults carry weight because most users never touch them, so Replit's spending on Claude is likely to fall even with the option intact. Masad said Replit ranked among Anthropic's five biggest customers until recently. Insiders familiar with the relationship disputed his onstage framing, and one said the partnership was intact. The move follows Enki's report that Meta and Microsoft have scaled back internal use of Claude.

Masad gave no ARR figure but sketched the revenue mix. Enterprise contracts grew from roughly 2% of revenue last year to about 15% this year, more than half of all revenue is now B2B, and he expects enterprise and government buyers to supply the majority in 2027.

What we don't know yet

Whether Replit's commercial agreement with Anthropic has changed, or only its routing.
Which models the automatic routing in Free Mode now favours for most requests.
How Replit's benchmark is built, and whether its results hold for work beyond new apps.
Whether Claude pricing or availability inside Replit will shift soon.