OpenAI shipped plugins for Codex on March 26, 2026, and the feature is real infrastructure — but it doesn't do anything Claude Code couldn't already do five months earlier, as SiliconAngle first reported.
The Codex CLI 0.117.0 changelog makes plugins a first-class workflow: Codex can sync product-scoped plugins at startup, browse them in a /plugins directory, and install or remove them with explicit auth handling. The OpenAI developers page describes a plugin as a bundle of skills (reusable prompts with optional scripts), app integrations, and MCP servers, packaged for team sharing. More than 20 plugins are available at launch across the Codex app, CLI, and VS Code extension, including Figma, Notion, Gmail, Google Drive, and Slack.
That sounds like a meaningful expansion. It isn't.
"By and large, they do not enable anything that was not possible before," Ars Technica reported. "Power users could already introduce custom instructions, use MCP servers, and so on to create much of this functionality. But in this case, it is basically a one-click installation." The feature packages what ambitious developers were already assembling by hand. One-click convenience is not new capability.
The gap between Codex and Claude Code on plugin volume is the more telling detail. Anthropic's Claude Code, which shipped its plugin system around October 2025, now advertises 415 plugins and 2,811 agent skills in a community-maintained registry — numbers that shift regularly and reflect a volunteer ecosystem, not a curated storefront. OpenAI launched with more than 20. The disparity in available integrations reflects ecosystem maturity that will take time to close, and whether developers invest that time in a third-party platform remains an open question.
On pricing, the comparison requires context. ChatGPT Pro, which includes Codex access, runs $200 per month. Claude Code Max is a standalone subscription at $100 per month. One developer told ZDNet they preferred Claude Code partly because they could get their work done on the cheaper tier. The token-efficiency claim is real — Codex runs on GPT-5.3-Codex, which runs 25 percent faster and uses 48 percent fewer tokens than GPT-5.2-Codex for equivalent results — but efficiency gains don't eliminate a direct price comparison gap.
Developers who talk to other developers find many more Claude Code users than Codex users in programming circles, Ars Technica noted. That's not a product verdict — it's an adoption signal. Plugins don't automatically change it.
So what is the real value? OpenAI's own guidance says to start local with skills, then package as a plugin when you're ready to share across a team. An official Plugin Directory is coming soon. The feature is less about extending what Codex can do and more about making what power users already built distributable to teams that don't want to configure MCP servers by hand. Standardization, not capability.
There is one genuinely interesting signal buried in the coverage. SiliconAngle reported that OpenAI plans to merge Codex and ChatGPT into a single desktop superapp later in 2026 — combining its coding agent with its general-purpose assistant. As reported, plugins would be the infrastructure that makes a unified product credible: shared plugin format, shared skills, shared MCP integrations across both surfaces. If the merger happens, today's plugin system is the foundation layer. If it doesn't, Codex plugins remain a workflow nicety.
Anthropic, for its part, has already moved past pure coding agents. The company launched Claude Cowork this year — a non-coding productivity version of Claude that uses customer-created plugins to automate marketing, finance, and other workflows, according to SiliconAngle. That's a different product category. OpenAI is still catching up in plugins while Anthropic is building the next layer up.
The multi-agent workflow improvements in 0.117.0 are worth noting separately. Sub-agents now use readable path-based addresses like /root/agent_a with structured inter-agent messaging — per the changelog. That's concrete UX improvement for teams running coordinated agent pipelines. The app-server-backed TUI is also now enabled by default. These are the kind of quiet infrastructure investments that matter in practice even when they don't appear in announcements.
Codex plugins are real. They ship today, they work, and the plugin format is genuine OpenAI infrastructure. But they're five months behind Claude Code on the feature, substantially behind on available integrations, and the subscription that includes Codex costs twice what a Claude Code Max standalone subscription costs. The superapp merger is the angle worth watching — if OpenAI actually consolidates Codex and ChatGPT, the plugin ecosystem gets a meaningful second chance. Until then, this is a feature catch-up with no structural advantage.
OpenAI declined to comment beyond its public documentation.
