OpenAI Routes Around the Model Picker , Routing Becomes the Product

2. OpenAI Routes Around the Model Picker , Routing Becomes the Product

OpenAI shipped automatic cross-model delegation inside multi-agents v2, confirmed by engineer Eric Provencher on August 15, 2026. The update lets agents delegate tasks to any supported model, including Luna, without the user choosing. Greg Brockman reposted Provencher's note with the framing: "towards never having to manually select a model again." Provencher acknowledged the team spent time ensuring reliability before shipping.

The move repositions model selection from a user-facing control to an infrastructure-level routing decision inside OpenAI's own stack. That is a direct competitive threat to the growing category of third-party model routers, including tools like Martian and RouteLLM, whose value proposition is exactly this: pick the right model for the right task. If OpenAI owns routing natively, those tools lose their primary wedge against ChatGPT users. It also shifts leverage away from users who want to pin specific models. The reply thread shows that tension directly: users demanding GPT-4o access are now one routing decision away from being silently served something else, with no clear override. Anthropic and Google both surface model-selection UI prominently in Claude.ai and Gemini; OpenAI is betting that removing that choice is a feature, not a regression.

The broader pattern is consolidation of the AI stack upward. First OpenAI abstracted the prompt. Now it abstracts the model. The next move to watch: whether OpenAI exposes routing configuration to API customers or keeps it opaque. Developers building on the API who need deterministic model behavior have a legitimate planning problem if routing becomes a black box. How OpenAI handles that tension will signal whether this is a consumer-only simplification or a platform-level architectural shift.

Source: Greg Brockman on X