ChatGPT Now Has 2 Million Plugins: Ecosystem, Plus and API

Two million plugins change ChatGPT’s operating model

Two million plugins may sound like a victory tally, but the more important story is the change in scale. ChatGPT is becoming more than an interface to a model: it is turning into an environment for installable workflows, skills, and connections to external tools.

In OpenAI’s current plugin documentation, plugins are described as reusable packages that can combine skills with tool connections, including MCP. Installing one involves finding or selecting a plugin, connecting it, and authorizing it when necessary. It can then be used in a new chat with ChatGPT or Codex.

This differs noticeably from the earlier approach described in OpenAI’s original plugin announcement. Back then, enabled plugins and their usage instructions were added to the model context. The focus has now shifted from a standalone extension to a packaged workflow in which the model receives a set of available actions at once.

At the time of this news, ChatGPT limits cannot be reduced to a single number either. OpenAI’s plan documentation says they depend on the plan, model, task complexity, and where computation takes place. Current limits and reset times appear in the usage panel, while published estimates should not be treated as fixed caps.

Plus and the API solve different problems

The key practical takeaway is simple: a Plus subscription and direct API access cannot be compared solely by an assumed price per request. Plus provides interactive access under a fixed subscription model, while API usage is billed separately based on actual consumption and receives no credits from a subscription.

For an average user, Plus is likely more convenient because spending is predictable and limits are built into a ready-made interface. For an application developer, programmable control and measurable API consumption matter more. Using subscription limits as a proxy for a product’s unit economics quickly breaks the calculation.

The scale of 2 million also makes the plugin count itself a weak quality signal. I would first examine the package contents, authorization, available tools, and actual behavior when limits are reached. The ecosystem is already enormous, but its maturity will now be determined not by the catalog, but by how predictably everything works together.

We previously covered OpenAI’s rollout of Codex within the ChatGPT Android app and what it means for product-based AI access. That shift helps frame the broader debate over using ChatGPT subscriptions versus building on the API.