Insights6 min read← All articles

The Battle for the Default Model Slot

Replit kept Claude in its dropdown menu while switching its default model to ChatGPT — a single config change that drains the bulk of Anthropic’s revenue from a top-five account. As rival labs cross-distribute each other’s models, whichever layer controls the default setting allocates foundation-model revenue.

On 17 October, Replit’s CEO remarked onstage at TechCrunch Disrupt that the company was “no longer working with Anthropic.” He later clarified that users can still manually select Claude from the dropdown menu, while ChatGPT has taken over as the platform’s default model.

01Flipping the default drains the majority of traffic

Until recently, Replit ranked among Anthropic’s top five customers. Making ChatGPT the default leaves Claude available in the settings menu, yet the vast majority of users never touch default configurations — meaning Anthropic’s revenue from a top-five account drops precipitously overnight.

Replit cited “cost, partnership, and performance,” adding bluntly: “I think ChatGPT is better now.” Sources close to Anthropic noted that the commercial relationship remains active. Both statements hold true: the contract is still in place, while the bulk of the token volume has already moved elsewhere.

At the application layer, changing a single configuration line is enough to redirect most of a major customer’s API spend.

02Rival labs are plugging into each other’s distribution

That same week (21 October), OpenAI announced that Google Cloud enterprise customers can purchase OpenAI frontier models directly through Google Cloud Marketplace, with native integration into Google’s Antigravity and Gemini Agent to follow. Meanwhile, Gemini Agent — launched on 8 October — already routes tasks automatically to Anthropic’s Claude.

Three labs competing head-to-head on frontier models are now cross-distributing each other’s APIs. Between the end user and the underlying model sits a growing stack of intermediaries that control the default selection:

  1. End-user applications (Controls the default)Products like Replit: one config line dictates which model most users run
  2. Agent host platforms (Controls the default)Gemini Agent: routes tasks automatically across rival models
  3. Cloud marketplaces (Controls the default)Google Cloud Marketplace: unified enterprise billing across competing labs
  4. Model routers (Controls the default)OpenRouter (acquired by Stripe): single API dynamically routing across all providers
  5. Foundation models (Underlying model)OpenAI, Anthropic, Google, and open-weight models
The software layers sitting between users and models. The top four layers control default routing, leaving model vendors at the bottom competing to be selected.

03Customer concentration risk and multi-model leverage

When a frontier model can be hot-swapped with a single config change, revenue allocation shifts to whichever layer owns the user’s default setting. Applications optimize defaults for margin and latency; agent hosts route subtasks by capability; routers send traffic to whichever endpoint offers the best price-performance ratio that day.

For model labs preparing to go public, this creates acute customer concentration risk. Losing the default slot at a single top-five account can wipe out a substantial chunk of quarterly ARR — a vulnerability public-market investors will scrutinize closely as OpenAI and Anthropic move toward IPOs.

Conversely, for application builders, multi-model compatibility is direct commercial leverage. Replit could flip its default from Claude to ChatGPT overnight because its product architecture avoided lock-in to any single vendor.

04Four takeaways across the stack

  1. 01For application teamsMaintain multi-model hot-swapping from day one. Credible ability to move traffic is your strongest bargaining chip when negotiating API pricing.
  2. 02For routers and platformsAutomated routing is the core value proposition. Platforms that provably cut inference spend without hurting output quality will capture growing share of token volume.
  3. 03For model labsDefault slots must be re-won with every model generation. Each major release triggers a fresh round of default-model evaluations across top applications.
  4. 04For investorsWhen evaluating a model lab, check top-five customer concentration and switching friction. When evaluating an application, ask whether it can swap foundation models within 24 hours.

As frontier models converge in capability and standardize around compatible APIs, bargaining power shifts toward the application and routing layers that control default model selection.

Replit’s switch shows how model market share actually changes hands: no public breakup, no delisting — just a quiet change to the default setting that moves tens of millions of dollars in annual run-rate.