Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

OpenAI details Performance Factory, a Codex-powered system built to keep ChatGPT fast

OpenAI published a post describing the Performance Factory, an internal system of Codex-powered workflows that finds and fixes performance issues to keep ChatGPT responsive. The disclosure shows how model makers are turning their own coding agents inward, using them to maintain the infrastructure behind large-scale AI services.

Published

OpenAI on Nov. 19 published a technical post introducing the Performance Factory, an internal system it describes as a set of Codex-powered workflows that finds and fixes performance issues to help keep ChatGPT fast.

The name suggests a continuous operation rather than a one-off optimization sprint. The verbs OpenAI uses are "finds" and "fixes" — locate the bottleneck, then drive the repair, closing the loop instead of stopping at diagnosis.

Codex, OpenAI's coding-focused model line, is the engine inside those workflows. In this framing it is less a tool for outside developers than an internal maintenance crew, working on the systems that serve ChatGPT's users.

The post does not disclose latency figures, cost savings, or the volume of issues handled, so the practical scale of the effort remains unquantified from the outside.

The direction still matters. ChatGPT operates at a scale where small regressions in response time translate into both user frustration and higher compute bills; automating part of performance engineering is a way to attack that problem without simply adding headcount.

It also reflects a broader shift in coding agents, which are moving from writing product code to maintaining production systems. Putting an agent in the performance loop is a natural next step once the agent can read a codebase, propose a change, and test it.

For competitors, the signal is that model service competition now runs through inference efficiency, time-to-diagnosis, and engineering automation, not just benchmark scores.

What to watch: whether OpenAI publishes harder numbers — detection time, fix cycles, resource usage — that would let outsiders judge how much the factory actually produces.

Why it matters

If the approach holds up, performance engineering could shift from manual tuning toward automated loops, making inference cost and reliability a sharper axis of competition among model providers.

OpenAICodexInfrastructure
Back to realtime news

Nearby Updates

All

11/19, 15:01

Kling Image 4.0 ships native 8K generation and 2,000-character text rendering

Kuaishou's Kling has released Kling Image 4.0, an image model that generates native 8K output with detail that still holds up under heavy zoom, according to a QbitAI hands-on test. The update also adds up to 2,000 characters of text per image, more than 70 preset fonts, series of up to 10 linked images and long e-commerce detail pages.

11/19, 14:57

AURA 2.0 lets the model start the conversation in streaming AI

AURA 2.0 has been released with model-initiated conversations inside streaming AI, according to Blockchain News. The change moves part of the right to start an interaction from the user to the system, which reshapes both the product and its boundaries.

11/19, 17:50

ServiceNow report: 92% of Indian organisations hit by AI agent security incidents

A ServiceNow report says 92% of Indian organisations experienced an AI agent-related security incident in the past year, according to varindia. The figure puts agent security at the centre of enterprise deployment decisions, though the published summary leaves the survey’s sample size, industries and incident definitions undisclosed.

11/19, 13:58

NTT invests in Japanese e-commerce AI agent developer Stract

NTT has invested in Stract, a Japanese developer of AI agents for e-commerce, according to a Telecompaper report carried as a headline in a Google News aggregation feed. The item did not disclose the size of the investment, a valuation, or which NTT business units might use the technology.