OpenAI AI Updates: November 20, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. OpenAI Describes the Performance Factory, a System of Codex-Powered Workflows That Finds and Fixes ChatGPT Performance Issues
OpenAI published an engineering post on the Performance Factory, a system of Codex-powered workflows that finds and fixes performance issues to help keep ChatGPT fast. The post describes how OpenAI uses its own coding agent to run this work on the product. Source
2. OpenAI Publishes a Reference Architecture for Supervised Security Investigations With Daybreak Red
OpenAI released a guide, “Deploying Daybreak Red,” that lays out a reference architecture for supervised security investigations with Daybreak Red, its restricted tier for specialized cyber models. According to the description, the guide covers safety controls, setup and monitoring. Source
3. Codex CLI 0.171.0 Adds GPT-6.2 Sol and Astra to Model Catalogs and Turns On the TUI Workspace by Default
OpenAI released Codex CLI 0.171.0, which adds GPT-6.2 Sol and Astra to the model catalogs, with Sol as the default selection. The TUI workspace is now on by default, with searchable document previews, image previews and subagent transcripts alongside the main conversation. The release also adds codex exec --remote <ADDR> to start or resume noninteractive sessions on an existing app server, /rc and /remote commands to manage local remote control and show device pairing codes, configurable undo and redo shortcuts in the prompt composer, and custom MCP status line selection through /statusline. Source
4. OpenAI’s Second Internal Research Acceleration Report Puts Median Researcher Agent Usage at $2,400 of Tokens a Day
OpenAI published its second report on how deeply AI agents are embedded in developing new models, with the stated aim of documenting indicators of recursive self-improvement. According to The Decoder’s summary, the median researcher would have consumed $2,400 of tokens per day at API list prices in October, up 62 percent from September, and the top 10 percent reached $23,000 per day. Agents ran the equivalent of 5.7 workdays per researcher per calendar day, more than two thirds of tracked researchers ran at least four agents in parallel at peak, and internal agent use accounted for about 2 percent of total research compute. Source