OpenAI Codex August 20: Codex CLI 0.148.0 Ships /export and codex exec fork, Codex Crosses 15 Million Users, and Paid Usage Resets Reach $80 on the $200 Pro Plan

OpenAI Codex CLI v0.148.0 adds /export to Markdown, codex exec fork, and cost in /status

• August 18: v0.148.0 exports complete TUI conversations to Markdown via /export, adds codex exec fork with archive and restore from the resume picker, shows estimated thread credits or cost in /status and terminal titles, adds Amazon Bedrock Runtime as a built-in provider with AWS profile, region and GPT-5.6 routing, and lets hooks run asynchronously and invoke MCP tools.
• Sandbox restrictions now fail closed for denied or unreadable paths across Linux and Windows.

Editor’s read: The fail-closed sandbox change is the one to read carefully. If your workflow depended on a denied path silently passing through, it will now stop.

OpenAI Codex comes to Linux with a full desktop app for Ubuntu, Debian and Fedora

• August 11: the Codex desktop app in ChatGPT is now available on Ubuntu, Debian and Fedora.

OpenAI Codex adds import from other agents for projects, chats, skills and plugins

• August 11: Codex and ChatGPT Work can now import projects, chats, skills and plugins from other agents.

Editor’s read: Skill portability across agents is becoming table stakes. If you maintain skills, expect them to be consumed outside the tool you wrote them for.

OpenAI DevDay Exchange goes global with developer events in October

• August 18: DevDay Exchange events are scheduled for Bengaluru, Tokyo, Seoul, Berlin, London and more this October.

OpenAI Codex usage: paid usage resets in testing at $5 to $8 on Plus and up to $80 on the $200 Pro plan

• August 18: paid usage resets are being tested, reported at roughly $5 to $8 on Plus and up to $80 on the $200 Pro plan.

OpenAI Codex crosses 15 million users as another usage reset ships

• August 13: Codex passed 15 million users, with Tibo Sottiaux shipping a further usage reset.

Z.ai GLM-5.3 targets OpenAI Codex and Claude Code users directly

• August 14: Z.ai positions GLM-5.3 against Codex and Claude Code, claiming 50 percent better coding than GLM-5.2 and the top Terminal Bench 3.0 score among open models, with weights to follow.

Theo: gpt-5.6-sol builds iOS apps with a dedicated Mac, computer use and the simulator

• August 11: Theo described gpt-5.6-sol building iOS applications end to end using a dedicated Mac, computer use and the iOS simulator.

swyx wraps the $10,000 Kill My SaaS hackathon with vibe-coded replacements for a $40K per year SaaS

• August 13: the hackathon asked developers to rebuild a SaaS product costing $40,000 per year, with a $10,000 prize pool.

Arvid Kahl on why developers stopped reading AI-generated code

• August 12: Kahl argues the test suite has quietly taken over the review job that developers used to do by reading diffs.

Arvid Kahl: tooling to guard agentic coding work is still missing

• August 11: Kahl points out that safety tooling around autonomous agent work has not kept pace with the agents themselves.