Replit is leaning towards OpenAI in its current default model preference and retaining the Claude option. The company’s multi-model agent approach demonstrates the difference between this choice and removing all alternatives. The way Replit works with AI agents is broader than just the model name that appears on the chat screen.
The default preference tells which provider the service goes to when the user does not select another model. On the other hand, the fact that a model is among the options means that the user can choose it separately. Therefore, OpenAI’s coming to the fore does not mean the same thing as Anthropic models leaving the platform completely.
The technical article published by Replit on September 29 discusses the main agent’s selection of assistant agents and thinking levels depending on the task. The company aims to use the resources of the powerful model in more difficult decisions by transferring routine tasks to lower-cost auxiliary models. The evaluation here is different from measuring the capabilities of a single model.
In addition, the same article gives examples of different Claude and OpenAI models in the production environment. The frequency of delegating work to subordinate agents and the behavior of returning to a previously assigned agent vary across models. Replit explains that it has established its working order to take these differences into account.
Replit shares tasks between different models
Free Mode, which the company announced in August, is another part of the OpenAI collaboration. This mode handles casual chatting, idea development and planning within the usage limits of the subscription. A more complex application development job may require different working modes.
On the other hand, choosing a vendor does not prove that a single model will provide the best results for all tasks. Designing an interface, finding the error in the current project, and connecting the data source require different skills. For this reason, the task distribution of the service becomes as important a decision as the model selection.
The model selection documentation explains that Core and Pro users can switch the primary model in Power or Max mode. Options on the Power list include the GPT-6 Sol, GPT-6 Luna Fast and Claude Sonnet models. In Max mode, besides GPT-6 Astra, there is Claude Fable 5.1, Opus options and Kimi K3.
According to the same document, Free Mode uses automatic model orientation and does not allow manual model selection. Additionally, options may vary depending on the deployment stage, organizational settings and account authority. The model list in the user interface shows the available options for that account. Thus, the company’s general provider preference and the models that an account can access remain separate information.