Codex Head Tibo Reveals Production Traffic Distribution
Tibo, head of OpenAI Codex, stated that currently about 5% of production traffic runs on Pi harness, and another 5% on OpenCode.
He reminded users that they can work with their ChatGPT accounts on other thriving tools and emphasized that Codex will continue to be optimized, while users have multiple options.
Internally, OpenAI accelerates the iteration of agentic tools through multi-path testing and traffic diversion, with developers tending to adopt flexible multi-tool combinations, while users relying solely on Codex face increased options, with funding flowing towards AI ecosystem platforms that support ChatGPT account interoperability.
Source: Public information
ABAB AI Insight
Tibo, as the head of the Codex project, previously promoted internal dogfooding (self-use iteration). This public disclosure of traffic distribution data continues OpenAI's consistent transparency style, similar to the early parallel testing paths of the GPT series, aimed at validating the actual performance of different agent execution frameworks through Pi harness and OpenCode.
On the capital path, OpenAI is mobilizing user base and data resources towards a multi-product matrix by allowing ChatGPT accounts to be cross-tool compatible, motivated by reducing the risk of reliance on a single Codex and enhancing overall ecosystem stickiness. Strategically, this forms a developer experience of "one account, multiple tool legion," accelerating the transition from code generation to full-stack autonomous agents.
Similar to OpenAI's previous maintenance of multiple internal experimental projects, Codex is currently in a phase of expanding from a core product to an open ecosystem, with the 5%+5% production traffic proving the maturity of alternative paths.
Essentially, this is a reconstruction of the industry chain driven by technological substitution. OpenAI is changing developers' dependence on pricing power of a single AI coding platform through multi-tool parallelism and account interoperability, with the mechanism being internal traffic diversion accelerating iteration, while shifting users from "locked tools" to "flexible combinations," prompting capital to concentrate from vertical single products to a horizontal AI agent ecosystem.
ABAB News · Cognitive Law
The earlier the traffic diversion, the lower the risk of a single product.
One account to rule them all is the key to ecosystem victory.
The best tools are never singular, but options that can be switched at any time.