OpenAI Codex Is No Longer Just for Coding: What Changed in 2026
If you still think of OpenAI Codex as the GitHub Copilot competitor from 2021, you haven’t been paying attention. In 2026, Codex quietly became an AI productivity tool that touches reports, spreadsheets, presentations, contracts, research, and full-blown enterprise workflows. On June 2, 2026, OpenAI released a report called The Next Era of Knowledge Work showing that Codex now has more than 5 million weekly active users — up more than 6x since the desktop app launched in February. Roughly 20% of those users are not developers at all. They’re analysts, marketers, designers, investors, and bankers.
I dug into the official announcements, the new model card, and the third-party reporting to figure out what actually changed. Here’s the honest, source-backed breakdown.
Pull quote: “Knowledge workers now represent about 20 percent of Codex users and are adopting it more than three times as fast as developers.” — OpenAI, The Next Era of Knowledge Work, June 2, 2026
What “Codex” even means in 2026
OpenAI Codex started in 2021 as a code-completion model. It then became a CLI coding agent. In 2026, “Codex” is the umbrella name for three things at once: a family of specialized models (GPT-5.3-Codex and friends), a cross-surface product (app, CLI, IDE, mobile, web), and a marketplace of plugins that connect it to the tools your team already pays for. That last layer is the part most people missed.
On June 2, 2026, OpenAI shipped the biggest expansion yet. Let me walk through what changed, what’s real, and what’s marketing.
The headline numbers (verified across three sources)
Before the details, here are the data points I could cross-verify across OpenAI’s own announcements, Constellation Research, and Help Net Security — all dated within the last three weeks:
- 5M+ weekly active users as of June 2, 2026 (OpenAI, Constellation Research, Help Net Security)
- 6x growth since the Codex desktop app launched in February 2026
- 20% of users are knowledge workers (not developers), and they’re growing 3x faster than developers
- 50% of users now run multiple Codex tasks in parallel, up from below one-third in mid-April 2026
- 72% of knowledge-worker users produce artifacts weekly (reports, memos, spreadsheets, presentations, contracts)
- Data-analysis usage is growing 110% week over week among knowledge workers
That’s not a coding tool anymore. That’s a productivity platform with a coding tool attached.
The model underneath: GPT-5.3-Codex
Everything starts with the model. GPT-5.3-Codex dropped on February 5, 2026 and is the first Codex model OpenAI classifies as “High capability” for cybersecurity tasks under its own Preparedness Framework (OpenAI). It’s also 25% faster than its predecessor and — this is the bit I care about — it goes “beyond coding.”
Here’s how it scores on the four benchmarks OpenAI itself publishes:
| Benchmark | GPT-5.3-Codex (xhigh) | GPT-5.2-Codex (xhigh) | What it measures |
|---|---|---|---|
| SWE-Bench Pro (Public) | 56.8% | 56.4% | Real-world software engineering across 4 languages |
| Terminal-Bench 2.0 | 77.3% | 64.0% | Terminal skills for coding agents |
| OSWorld-Verified | 64.7% | 38.2% | Computer-use tasks on a real desktop (humans score ~72%) |
| GDPval (wins or ties) | 70.9% | — | Professional knowledge work across 44 occupations |
| Cybersecurity CTF | 77.6% | 67.4% | Capture-the-flag security challenges |
OSWorld-Verified is the one to watch. It tests whether the model can drive a real desktop — clicking, typing, opening apps, reading screens. Codex jumped from 38% to 64.7% in one model generation. That’s the difference between “occasionally useful” and “I can hand it a spreadsheet and a PDF and walk away.”
OpenAI’s own description is striking: “With GPT-5.3-Codex, Codex goes from an agent that can write and review code to an agent that can do nearly anything developers and professionals can do on a computer.” (OpenAI, Feb 5, 2026)
Six role-specific plugins that change who Codex is for
The single biggest 2026 shift isn’t the model. It’s the plugins announced on June 2. Together they bundle 62 popular apps and 110 automated skills into ready-made packages. You don’t write code. You install a plugin and tell Codex what you need.
The six plugins that shipped on day one:
- Data Analytics — Snowflake, Databricks Genie, Hex, Tableau. Ask “why did churn spike last quarter” and it queries your warehouse, drafts a chart, and explains the movement.
- Creative Production — Figma, Canva, Shutterstock, Picsart, Fal. Turn a creative brief into display ad variations, campaign boards, or product lifestyle shots.
- Sales — Salesforce, HubSpot, Slack, Outreach, Clay, Rox, Actively. Surface high-priority accounts, prep meeting briefs, update CRMs, build close plans.
- Product Design — Figma and Canva. Audit live user flows, prototype from a URL, turn static screenshots into interactive demos.
- Public Equity Investing — Moody’s, Daloopa, Datasite, FactSet, LSEG, S&P, PitchBook, Hebbia. Review earnings, compare companies, stress-test an investment thesis.
- Investment Banking — Trusted market data to turn diligence into client-ready pitch decks and comp tables.
More are coming: Corporate Finance, Private Equity Investing, Marketing Strategy, Strategy Consulting, and Legal. OpenAI is also opening the door to third-party plugins — they’re explicitly building toward a partner ecosystem the way the ChatGPT plugin store tried to in 2023, only this time with a much sharper focus on white-collar work.
This is the part where Anthropic’s Claude Cowork started nipping at OpenAI’s heels earlier in 2026. The New Stack called out the overlap directly: “This list of verticals sounds awfully similar to the plugin collection Anthropic released for Claude Cowork earlier this year” (The New Stack, June 2, 2026). That’s the polite version of: the two companies are now in a direct feature war for the same knowledge-worker budget.
Sites: the feature that made me do a double take
Codex Sites is the new bit I didn’t expect. Announced in preview on June 2, Sites lets Codex generate interactive, hosted web apps and share them via a URL inside your workspace.
The use case OpenAI demos: turn a financial model into a scenario planner your CFO can poke at in a browser. Or turn launch materials into a living hub with the latest messaging, milestones, and decisions. Or build a customer-review site that pulls in usage trends and open questions for a specific account.
VentureBeat’s Carl Franzen explains the technical bit: Sites are hosted by OpenAI, not generated as static files you export. They’re rolling out first to Business and Enterprise customers through the Codex app. Partners helping OpenAI build toward a Sites ecosystem include Vercel, Wix, Base44, Replit, Lovable, Figma, Webflow, and Emergent.
Annotations: the small feature that kills a giant annoyance
Before annotations, asking an AI to “add a chart of revenue, EBITDA, and net income” in a financial model meant the model rewrote the whole file. That broke custom formatting and lost dependencies.
Annotations fix this by letting you highlight a section of a document, spreadsheet, slide, or site and tell Codex exactly what to do with just that slice. The model maps the underlying schema, isolates the selected data array, and edits only that boundary (VentureBeat, June 2, 2026). Developers already had this for code; the June update extends it to the documents, slides, and sheets that knowledge workers actually live in.
The corporate reorg that made it inevitable
You can’t ship all this without an org chart to match. On May 15, 2026, Greg Brockman was officially named permanent head of product strategy at OpenAI. In an internal memo seen by Wired, he wrote that OpenAI would “merge ChatGPT and Codex into one unified experience.” Three product lines — ChatGPT, Codex, and the developer-facing API — are now one team.
Two notable moves came with it:
- Thibault Sottiaux, the engineer who built Codex into one of OpenAI’s fastest-growing products, now leads core product and platform across consumer, enterprise, and developer surfaces.
- Nick Turley, who grew ChatGPT to 900M+ weekly users, moves to enterprise products.
The Next Web’s May 17, 2026 writeup reads the move as OpenAI “killing side projects to focus on one agentic platform before its IPO.” Whether or not the IPO lands this year, the signal is clear: Codex isn’t a developer tool that may someday expand. It’s the kernel of OpenAI’s entire product strategy.
Codex vs. the field: where it actually stands
Anyone comparing AI coding tools right now is juggling at least three names: Codex, GitHub Copilot, and Claude Code. Here’s how I read the 2026 landscape:
| Tool | Best at | Starting price | Weekly users / scale signal |
|---|---|---|---|
| OpenAI Codex | End-to-end agent work across coding, docs, spreadsheets, sites | Free / $20 Plus / $100 Pro / Business | 5M+ WAU as of June 2, 2026 |
| GitHub Copilot | Inline IDE completion, lightweight agent mode | $10/mo Pro / $19/mo Business | Largest installed base; mature IDE integrations |
| Claude Code | Long-running terminal coding sessions, deep repo reasoning | $20/mo Pro; enterprise pricing varies | Anthropic declined to share WAU publicly |
The honest 2026 take from the data I read: Copilot wins on cheap, frictionless IDE completion. Codex wins on cross-surface agent work (it lives in your IDE, terminal, desktop app, browser, and phone, and now in your docs and sheets). Claude Code wins on raw coding taste in benchmarks and in long-running refactors where its reasoning holds up.
If your job is “ship features in a VS Code workflow,” Copilot is still excellent. If your job is “automate the messy, cross-tool work my team does all day,” Codex’s 2026 expansion is the most direct answer I’ve seen.
Who’s actually using it
Help Net Security’s June 2 breakdown of OpenAI’s report surfaces three case studies worth quoting:
- GroundVue uses Codex to make public meetings from roughly 90,000 government bodies searchable. Tasks that took days now take minutes.
- Proaction, a five-person fleet-management startup, turns customer conversations into proposals and working demos before contracts are signed.
- Taiyo Inoue, a math professor at Cal State, uses Codex to maintain his Canvas courses and estimates it saves him 4–5 hours a week.
None of these people write code for a living. That’s the point.
What it costs in 2026
Codex is bundled with every ChatGPT tier. Pricing as of June 2026 (chatgpt.com/codex/pricing, OpenAI’s April update):
- Free — $0, limited Codex usage
- Go — $8/month
- Plus — $20/month, the practical starting point
- Pro — $100 (5x) or $200 (20x) per month
- Business — $20/seat/month annual, with pay-as-you-go credits
- Enterprise — custom, includes Sites and admin controls
Sites and the role-specific plugins are Business and Enterprise only at launch. IT can lock down which apps a Codex agent can touch through workspace settings.
What this actually means for you
If you’re a builder or operator trying to figure out whether to care, here’s my honest read after going through all this:
- If you write code: GPT-5.3-Codex is a real upgrade, especially for long-running tasks. The 25% speed bump matters. The computer-use jump (38% → 64.7% on OSWorld-Verified) means Codex can drive a browser or a desktop app on your behalf, which it couldn’t reliably do six months ago.
- If you’re an analyst, marketer, operator, or founder: Codex in mid-2026 is the first AI tool that genuinely does the whole loop — pull data, build the deck, ship the site, file the CRM update — without you babysitting every step. Try the data analytics plugin before you sign up for another BI seat.
- If you lead an enterprise team: the June 2, 2026 OpenAI announcements and the Frontier platform (announced April 8 with Oracle, State Farm, and Uber as anchor customers) are the parts to evaluate. The plugin directory, the Sites preview, and the merged ChatGPT-Codex-API roadmap under Brockman are the structural story.
The bottom line
OpenAI Codex in 2026 is no longer a coding tool that does knowledge work on the side. It’s a knowledge-work tool that started in coding. The 5M weekly active users, the 6x growth since February, and the 3x faster adoption among non-developers tell the same story three different ways. Whether you trust the hype is up to you, but the underlying shift — that the model that writes your code can also build your dashboard, draft your pitch deck, host your scenario planner, and update your CRM — is real, dated, and verifiable.
I’d still verify any single claim before betting budget on it. But the direction isn’t ambiguous anymore.
Primary sources cited in this article:
- OpenAI, “Codex is becoming a productivity tool for everyone” — openai.com/index/codex-for-knowledge-work (June 2, 2026)
- OpenAI, “Codex for every role, tool, and workflow” — openai.com/index/codex-for-every-role-tool-workflow (June 2, 2026)
- OpenAI, “Introducing GPT-5.3-Codex” — openai.com/index/introducing-gpt-5-3-codex (Feb 5, 2026)
- OpenAI, “The next phase of enterprise AI” — openai.com/index/next-phase-of-enterprise-ai (April 8, 2026)
- Wired, “Greg Brockman Officially Takes Control of OpenAI’s Products” — wired.com/story/openai-reorg-greg-brockman-product (May 15, 2026)
- VentureBeat, “OpenAI’s Codex update lets agents build interactive enterprise workspaces” — venturebeat.com (June 2, 2026)
- The New Stack, “OpenAI’s Codex adds new tools — Sites, Annotations, more plugins” — thenewstack.io/openai-codex-knowledge-workers (June 2, 2026)
- Help Net Security, “Codex knowledge work expands into research, reports, and spreadsheets” — helpnetsecurity.com (June 2, 2026)
- Constellation Research, “OpenAI touts broadening Codex usage with 5 million weekly active users” — constellationr.com (June 2, 2026)
- The Next Web, “OpenAI merges ChatGPT and Codex under Greg Brockman” — thenextweb.com (May 17, 2026)