The Replit default model for Agent is now supplied by OpenAI, chief executive Amjad Masad said onstage at TechCrunch Disrupt on 16 October 2026. He told the audience Replit is no longer working with Anthropic, then clarified afterwards that Claude models remain selectable. Replit’s changelog carries no entry announcing the change.
What Replit’s Chief Executive Said
Masad made two statements a day apart, and the second one softened the first considerably. Both came through TechCrunch, which ran the interview at its own conference.
Onstage on Thursday 16 October 2026, Masad — co-founder and chief executive of Replit, the San Francisco browser-based coding platform behind Replit Agent — said the company is no longer working with Anthropic, the San Francisco AI company that makes Claude. He gave cost as one reason among several, and on performance said simply, “I think ChatGPT is better now”, citing internal and public benchmarks on building applications from prompts. He described the decision as based on “both cost, partnership, and performance”.
Speaking to TechCrunch afterwards, he narrowed it. “Replit remains model agnostic,” he said, and “OpenAI is now our preferred partner for powering Replit Agent.” Customers can still select Anthropic’s models. Masad also said Replit had recently been one of Anthropic’s top five customers, which is why the change of default matters commercially: most users never open the model picker at all.
What Replit’s Changelog Shows Instead
Replit’s own product documentation does not record a default-model change, and the entry published on the day Masad spoke added an Anthropic model rather than removing one. The sequence below comes from Replit’s dated changelog pages.
| Changelog date | What it says about models |
|---|---|
| 28 August 2026 | Intelligent Model Routing becomes the default Agent mode for Enterprise workspaces; admins can restrict which models are available |
| 2 October 2026 | “You can now choose GPT-6.1 Sol and Claude Sonnet 5.5 when building with Agent” — Sonnet 5.5 in Power Mode, GPT-6.1 Sol in Max Mode |
| 9 October 2026 | Windows desktop preview, cross-project chat, TikTok Ads; no mention of models or defaults |
| 16 October 2026 | “Builders on Core and Pro can now choose Claude Haiku 5.5 from the model selector in Power Mode” |
That last entry is dated the same day as Masad’s remarks. The words “OpenAI” and “Anthropic” do not appear in it, and neither does any reference to a default.
Replit’s Agent documentation is similarly silent. It describes model selection in terms of modes — Auto, Power, Max and Free — and states that “Auto chooses the mode and an allowed model for each task” in Enterprise workspaces. It names no model and identifies no default anywhere on the page.
Why the Accounts Conflict
The conflict is between what an executive said at a conference and what the company has written down, and it has not been resolved. Three specific points remain open.
- Whether the partnership ended: Masad said Replit is no longer working with Anthropic. A person familiar with the arrangement told TechCrunch it remains in place. Anthropic has not commented.
- Which model is the default: Masad said ChatGPT. Neither he nor TechCrunch named a specific model, and Replit’s documentation names none. Secondary blogs circulating the names GPT-6 Sol and GPT-6 Astra are not supported by anything Replit has published.
- When the switch took effect: No date has been given, and no changelog entry marks it. A user cannot establish from Replit’s own records which model ran a session on any given day.
Replit also promoted Anthropic models on its site, blog and social accounts as recently as late September 2026, naming Claude Fable 5 and 5.1 and Opus 5.5, according to TechCrunch’s review of those posts. Nothing published before 16 October signalled a change.
What It Means for Projects Built on Replit Agent
If you build on Replit and have never touched the model selector, your next Agent session may run on a different model than your last one did. That is the whole practical consequence, and it is worth checking rather than assuming.
- Open the model selector: It is the only authoritative record of what your workspace can reach. Replit’s changelog confirms Claude Sonnet 5.5 and Claude Haiku 5.5 in Power Mode and GPT-6.1 Sol in Max Mode for Core and Pro builders.
- Choose a model rather than leaving it on Auto: Auto lets Replit pick, which is exactly the setting a default change moves.
- Re-test anything important: A different model can produce different code for the same prompt. Agent-built applications you depend on are worth running through their main paths again.
- Enterprise administrators can restrict models: Enterprise model controls, in place since the 28 August 2026 changelog, let an admin limit which models a workspace may use, and users can turn Auto off to pick an enabled model.
Replit has not announced any pricing change tied to the switch, and none appears in the changelog. The mode structure that governs cost — Free, Power and Max — is unchanged.
Why a Default Change Carries Commercial Weight
A default is worth more than a menu entry because almost nobody changes it. That is the mechanism behind Masad’s own framing of the decision as partly about cost.
By his account Replit was lately among Anthropic’s five largest customers, so moving the default — rather than removing Claude — shifts the bulk of inference spending without closing the door. Model pricing has been falling on both sides: API prices came down by up to 50% at OpenAI and 20% at Anthropic in September 2026. Anthropic’s finances are also now a matter of public record, with an operating loss of $8.1 billion disclosed in its IPO prospectus, which makes the loss of a top-five customer’s default traffic a figure analysts will look for. OpenAI, for its part, has been pushing its newest models to wider tiers, including GPT-6 reaching free users in October 2026.
Status as of 19 October 2026
As of 19 October 2026, OpenAI is Replit’s stated preferred partner for Agent, Anthropic’s Claude models remain selectable, and Replit has published no changelog entry or documentation describing a default-model change. Its most recent model-related entry, dated 16 October 2026, added Claude Haiku 5.5 to the selector.
Anthropic has not commented publicly. Whether the commercial partnership has ended, as Masad said onstage, or continues, as a person familiar with it told TechCrunch, has not been settled by either company.
Frequently Asked Questions
Can I Still Use Claude on Replit?
Yes. Masad said customers can still select Anthropic models, and Replit’s changelog of 16 October 2026 added Claude Haiku 5.5 to the model selector for Core and Pro builders in Power Mode.
Which Model Is the Replit Default Model Now?
Replit has not named one. Masad said ChatGPT is now the default, but neither he nor Replit’s documentation identifies a specific OpenAI model, and the Agent documentation names no default at all.
Has Replit Ended Its Anthropic Partnership?
Accounts differ. Masad said onstage that Replit is no longer working with Anthropic; a person familiar with the partnership told TechCrunch it remains in place. Anthropic has not commented.
Will My Existing Replit Apps Change?
Code already written does not change. Future Agent sessions may run on a different model if you leave the selector on Auto, which can produce different output for the same prompt.
Does the Switch Change What Replit Costs?
No change has been announced. Replit’s Free, Power and Max mode structure is unchanged in the changelog, and no pricing update accompanied the switch.
