Replit Makes OpenAI Its Default Over Anthropic, CEO Says

Introduction

On Thursday, October 15, 2026, Replit CEO and co-founder Amjad Masad said onstage at TechCrunch Disrupt (the day per TechCrunch's report) that Replit is "no longer working with Anthropic," citing several reasons including cost. TechCrunch's report was later updated and re-titled after Masad clarified the substance: OpenAI is replacing Anthropic as Replit's default model provider, and customers can still access Anthropic's Claude models on the platform. The updated headline reads "Replit makes OpenAI its default over Anthropic, CEO says."

The accurate story is therefore a change of default, not a removal of Claude. This review looks at what that means for Replit users, using Replit's own documentation and blog posts to describe how model selection currently works.

What was said and what was clarified

TechCrunch reports that Replit announced a partnership with OpenAI in August, and that before that Replit had relied on Anthropic's Claude models to run its platform. Masad described the decision as based on cost, partnership and performance, and said an internal benchmark helps assess performance. Afterward he told TechCrunch: "Replit remains model agnostic... OpenAI is now our preferred partner for powering Replit Agent, based on performance, economics, and the strength of our partnership."

TechCrunch also reports that one person familiar with the Anthropic-Replit relationship, speaking anonymously, disputed the onstage remarks and said the partnership was still in place. That account is unconfirmed, and TechCrunch notes that Replit had been promoting Anthropic products within recent weeks. We treat the default change as confirmed by Replit's CEO and the broader contractual picture as unsettled.

Feature Overview

1. Auto: intelligent model routing

Per Replit's model selector documentation, Auto means Replit is performing intelligent model routing. The router considers the requested work and its complexity: a focused request can use a lower-cost model, while complex work may need a more capable one, and Replit can change the model as the task changes.

2. Manual choice in Power and Max modes

Builders on Core and Pro can choose a model themselves in Power Mode or Max Mode. The documentation lists the following:

The list shows both providers side by side, which is consistent with Masad's "model agnostic" framing. Replit's pages are not perfectly consistent on naming: the table above uses GPT-6 names, while the August Free Mode post refers to OpenAI's GPT-5.6 Luna.

3. Effort control and model comparison

Effort is a per-model control from Low to Max that sets how much reasoning the selected model spends on each response. Replit says higher effort invokes its most capable frontier models, applied selectively on genuinely hard requests. A Compare models feature, in beta, runs the next prompt across up to four models so users can judge results side by side.

4. Free Mode on an OpenAI model

In its August 18, 2026 launch post, Replit said Free Mode is powered by OpenAI's GPT-5.6 Luna. Core subscribers, at $20 per month, can create up to 30 times more than before, plus up to 30 hours per month of chat. Core and Pro users can use Free Mode until they reach usage limits, which reset every 5 hours. Power Mode (formerly Economy Mode) and Max Mode remain available.

5. Max mode with GPT-6 Astra as the core loop

In "Free the models," published September 29, 2026, Replit says Agent in Max mode uses GPT-6 Astra as its core loop, which chooses subagent tiers and effort as the task unfolds. The post reports the following results, which are Replit's own self-reported numbers:

Replit's reading is that neither Astra baseline wins on both cost and score. The same table shows the trade-off clearly: Astra alone scores higher at its best setting, but at more than twice the cost per task.

Usability Analysis

For most users the visible change is the default. TechCrunch notes that most customers stick with the default rather than choosing a model, so a default switch can shift model spend more than any change to the selectable list. Users who prefer a particular Claude model can still pick it in Power or Max mode if they are on Core or Pro.

The documentation shows control at several levels. Individuals choose a model and effort. Teams can run the same prompt across up to four models. On Enterprise, account admins use model controls to set which providers and models are available in each Workspace, which matters for organizations with vendor policies.

A practical caution applies to cost. Replit's blog frames Free Mode as not consuming credits for everyday tasks, while Power and Max modes use usage differently. Users comparing models should check their own usage rather than assume the benchmark costs per task transfer to their projects.

Pros and Cons

Strengths: the Claude lineup remains selectable, the documentation is specific about which models sit in which mode, and Replit publishes cost-per-task alongside scores instead of scores alone. Effort control and Compare models give users ways to test rather than trust a default.

Limitations: the benchmark figures come from Replit and describe its own harness, so they are not independent verification. The reasons Masad gave beyond cost are general (performance, partnership), and no contract terms are public. The report itself notes a conflicting account from an anonymous source, and the extent of the change for existing users depends on how much they rely on the default.

Outlook

TechCrunch reports that Masad said on stage Replit had been, until recently, a top-five customer of Anthropic. If the default shift takes hold, model-provider mix on Replit will depend on defaults and Auto routing as much as on explicit user choice. Watch for changes to the Auto router, to the model list in the documentation, and for any clarification from Anthropic.

More broadly, the episode shows how a platform that supports many models can change its default quickly, and how that default, rather than the full menu, determines most usage.

Conclusion

Replit's change is best read as a default-provider switch toward OpenAI, announced in onstage remarks that were initially phrased more broadly than the CEO's later clarification. Claude remains available. Developers who rely on a specific model should pin it manually; teams should check Workspace controls; and everyone should treat the vendor-reported benchmarks as one input. This is a moderately significant tools story with some unresolved details.

Editor's Verdict

Replit's switch to OpenAI as its default model provider is a meaningful change for its users, even if it doesn't fundamentally change the landscape.

The strongest case for paying attention: Claude models remain selectable in Power and Max modes, so users keep a choice. That alone raises the bar for what readers should expect in this space. Reinforcing that, a documented model list by mode and an effort slider give users clear controls — practical value rather than just headline appeal. The broader signal worth registering is straightforward: the story is a change of default provider, not the removal of Claude, which Replit says remains available to customers. On the other side of the ledger, one constraint is real rather than a marketing footnote: benchmark results are self-reported by Replit and have not been independently confirmed in the cited sources. It should factor into any serious decision. Layered on top of that, the CEO's initial onstage wording was broader than the later clarification, which can confuse users and customers — which narrows the set of teams for whom this is an obvious yes.

For Replit users who rely on Auto or the default model, the smart move is to check which model is running and pin a specific one if results matter. For everyone else, the safer posture is to watch for Anthropic's response and further changes to Replit's model list.