Replit Makes OpenAI Its Preferred Partner—but Claude Remains Selectable

In TechCrunch’s October 17 update, Replit CEO Amjad Masad described OpenAI as the “preferred partner for powering Replit Agent” and clarified that users can still choose Anthropic’s Claude or other supported models. OpenAI is now Agent’s default provider. The clarification followed a broader onstage statement about Replit’s relationship with Anthropic that had suggested a complete break.

For an eligible Agent user, the distinction is consequential: leaving model choice to Replit now favors OpenAI, while manually selecting a supported Claude model remains possible. Free Mode has a separate OpenAI-powered setup and no manual model choice. The status of any commercial agreement between Replit and Anthropic does not, by itself, determine which models appear in Agent’s selector.

What a preferred partner changes in Agent

A preferred partner is the provider Replit has chosen to emphasize for Agent. A default is what a user receives without selecting a model. Neither designation makes the arrangement exclusive: Claude can remain an option even as OpenAI becomes the provider Replit favors for the standard experience. That is the practical boundary of the October clarification.

The earlier onstage description concerned the relationship between the companies; the later clarification addressed what customers can use. A selectable Claude model establishes continued product access, but it does not reveal whether the companies retain special pricing, joint development commitments or another commercial arrangement. Nor does the preferred designation disclose how much Agent work reaches each provider. Those are different questions from whether a customer can choose Claude.

The distinction also matters when interpreting “default.” Agent can use intelligent routing, choosing a model according to the task and changing models as work progresses. A provider preference therefore does not identify a single OpenAI model that runs every Agent request. A manual primary-model selection is a separate control available in eligible paid modes.

Free Mode and Agent have different model choices

Replit’s model availability is easier to understand by separating the mode that routes automatically from the modes that expose a selector. The published Agent catalog includes models from several providers, while an individual account may show a narrower set during rollout or because of organization settings.

• Free Mode: OpenAI’s August 19 account identifies GPT-5.6 Luna as powering Free Mode and describes routing a task that needs more advanced reasoning to GPT-5.6 Sol. Free Mode selects models automatically; users cannot manually choose Claude as its primary model.
• Agent with the default choice: OpenAI is the preferred provider when users leave model choice to Replit. That provider-level preference does not promise that every request uses the same model; intelligent routing can respond to the complexity of the work.
• Agent with manual choice: The Zevin Intelligence Journal’s October 17 documentation check counted seven Claude models, three OpenAI models and Kimi K3 among eleven models listed for selection in Power Mode and Max Mode. Core and Pro users can choose an available primary model in those modes, subject to what their accounts enable.

That catalog makes the difference between preference and removal concrete. Claude Sonnet models appeared in the Power Mode list, while the Max Mode list included Claude Fable and Opus models alongside other providers. Some Fast variants require a higher plan, and model selection is still rolling out. The list visible in a user’s own selector determines what that account can choose; a published catalog is not a guarantee that every listed option appears for everyone.

Free Mode’s OpenAI foundation predates the Agent partnership clarification. Its automatic routing also means that a named foundation model should not be read as a promise that every Free Mode task stays on that model. Conversely, the presence of Claude in Agent’s paid selector says nothing about a manual Claude choice in Free Mode, where the selector is unavailable.

Why Replit favors OpenAI

The stated rationale combines economics, the strength of the OpenAI relationship and performance on Replit’s app-building work. Model inference is a recurring cost for an agent that handles user requests, so provider economics can influence a default even when alternatives remain selectable. The earlier Free Mode deployment shows how lower model costs can widen access: OpenAI’s price-performance improvements helped make that mode viable. It does not disclose the supplier terms behind the later Agent preference.

The performance argument is narrower than a general ranking of coding models. It centers on a claim that Sol and Astra consistently did better at building applications from scratch in Replit’s internal evaluation and a public benchmark it publishes. That is Replit’s assessment of its own app-building task. The remark did not provide scores, test prompts or a method for independently comparing the results, so it cannot establish how the models would rank across every kind of development work.

An Agent result also reflects routing, available tools and the type of application requested. A model that performs well in one app-building evaluation need not be the preferred choice for every task or every user. Replit’s default expresses the provider it currently favors for its service; manual selection preserves a different choice where the account and mode allow it.

For someone who depends on a particular Claude model, the account-level selector remains decisive. Rollout and organization settings can narrow the published choices, even while Claude remains in Replit’s documented Agent catalog.

Also read:

• Personio Buys Circula—but the Price Remains Hidden
• Spotify Is Selling Its Internal Developer Stack—Four Tools Now Share One Home
• Claude Cyber Access Expands to Three Tiers—but Verification Still Gates It