OpenAI Codex in 2026: 2M Users, Astral Acquisition, and How It Actually Works

OpenAI Codex now has 2 million weekly active users — up 3x since January 2026, with 5x usage growth in the same period. On March 19 it acquired Astral, the startup behind uv and Ruff, the Python tools that millions of developers run every day. If you have not used Codex recently, it is a fundamentally different product than the code-completion tool from 2021.

Here is what Codex actually is in 2026, how it works, what it costs, and how it compares to Cursor and GitHub Copilot.

What Codex Is Now — Two Products, One Brand

Codex in 2026 is two separate but complementary tools under the same name.

Codex cloud agent — an agentic coding system that runs tasks asynchronously in isolated cloud containers. You give it a task ("fix the authentication bug in this repo", "write tests for the payment module", "create a PR that migrates this codebase from pip to uv"), and it works in the background while you do something else. Accessible at chatgpt.com/codex, the macOS app, and the Windows app (launched March 4, 2026).

Codex CLI — an open-source terminal-based coding agent. Install it with npm i -g @openai/codex or brew install --cask codex. It runs locally in your working directory, reads and edits files, and executes terminal commands. Built in Rust. Powers models including o3 and o4-mini.

The practical split: use the CLI for interactive local coding sessions where you want to stay in the terminal, use the cloud agent for batch or parallelizable tasks you want to run in the background.

What It Can Actually Do

This is where 2026 Codex is meaningfully different from what most people assume it is.

Full repository navigation. Codex reads your entire codebase, understands the file structure, and edits files across multiple paths in a single task. It explains its plan before making changes and lets you approve or reject steps.

Terminal command execution. The CLI runs arbitrary shell commands, tests, and build scripts locally. The cloud agent runs the same inside an isolated container with internet disabled during task execution — a deliberate security decision.

GitHub integration. Connect your GitHub account and Codex works directly with your repos. It can create pull requests from completed tasks and, if enabled, automatically review every new PR — it matches the PR intent against the actual diff, reasons over the full codebase, and runs tests to validate behavior.

Parallelization. You can spin up multiple Codex subagents to work on separate tasks simultaneously. Fix a bug, write tests for another module, and update documentation — all running in parallel while you review the first completed task.

Automatic code review. A separate Codex agent can review your code before you commit. This is distinct from the PR review feature — it is a pre-commit check you invoke manually.

The March 2026 changelog added GPT-5.4 mini as a faster, cheaper model option inside Codex (March 17), full-resolution image inspection in the CLI (March 16), a JavaScript REPL with filesystem helpers, and a themes system with custom colors and fonts (March 12).

Pricing

Via ChatGPT subscription:

• Plus ($20/month) — Codex included with a moderate usage cap (a few tasks per day)
• Pro ($200/month) — 6x higher limits, suited for full-time use
• Team ($25/user/month) — per-user limits similar to Plus

Via API (programmatic access):

• codex-mini-latest: $1.50 per million input tokens, $6.00 per million output tokens, with 75% prompt caching discount
• Full GPT-5: $1.25 input / $10.00 output per million tokens

For developers building applications on top of Codex, the codex-mini-latest model with caching is the cost-effective choice. The 75% caching discount is significant for agentic workflows that repeatedly reference the same codebase context.

The Astral Acquisition — What Changes for Codex

On March 19, OpenAI announced it will acquire Astral, the startup behind uv (Python package manager), Ruff (linter), and ty (type checker). The Astral team joins OpenAI specifically to work on Codex.

The gap this fills: Codex could already write functions, fix bugs, and run tests. What it could not do reliably was manage Python environments — the right Python version, dependency resolution, package installation. These are exactly the problems uv solves, and it solves them 10 to 100 times faster than pip.

With Astral's tools integrated into Codex, an agent that writes Python can also set up its own environment, install the right packages, run Ruff to lint its output, and verify types with ty — all in a single automated workflow. No human intervention needed to get the environment to a working state.

The integration is not live yet — the deal is pending regulatory approval. All three tools remain open source under existing licenses. But the roadmap is clear: Codex is moving from a tool that writes code to a tool that manages the entire development lifecycle.

For more detail on what the Astral acquisition means for developers using uv and Ruff today, see our OpenAI Astral acquisition breakdown.

Codex vs Cursor vs GitHub Copilot

These three tools are not direct competitors — they optimise for different workflows.

Cursor is an AI-first IDE (VS Code fork). It is best for active, real-time pair programming with multi-file edits. You are in the editor, you are the driver, Cursor is assisting. It benchmarks at 52% on SWE-bench and is 30% faster than standard coding workflows.

GitHub Copilot is an IDE extension across 6 editors. Its strongest advantage is GitHub ecosystem integration — issue-to-PR agents, enterprise SSO, and the fact that it already lives inside VS Code, JetBrains, and Neovim for most developers. It uses multiple models including Claude 3 Sonnet, Gemini 2.5 Pro, and GPT-5. At $10/month it is the cheapest option.

Codex is for autonomous and parallelizable tasks. The key word is async — you give it a task, it runs in the background, you get a PR or a result. You are not sitting in the editor watching it work. This makes it best for batch jobs: migrate this codebase, write tests for this module, review these 10 PRs. The cloud agent specifically shines for tasks that require running the code, not just writing it.

The developer consensus in 2026: use Cursor for active coding sessions, use Codex for background tasks that need full execution capability, use Copilot if you are in an enterprise environment already invested in GitHub.

Why 2 Million Users Is a Real Number

The 2 million weekly active users figure came from OpenAI at the Astral acquisition announcement. At the time, Codex had grown 3x in users and 5x in usage since January 2026 alone.

For context: GitHub Copilot hit 1 million users in its first year (2022–2023). Codex reached 2 million weekly active users within roughly 18 months of relaunch. The growth rate suggests it has crossed from "developer experiment" to "part of the daily workflow" for a meaningful segment of professional developers.

The usage growth (5x) outpacing user growth (3x) means existing users are doing more with it — more tasks per week, more complex tasks, more parallelization. That is the usage pattern of a tool that has become genuinely useful, not one that people try once and forget.

India is one of the fastest-growing markets for Codex. ChatGPT has 100 million weekly active users in India, with developers under 30 making up 80% of usage. Python dominates India's AI/ML, fintech, and SaaS sectors — which makes the Astral acquisition (uv + Ruff for Python) directly relevant to the Indian developer market. The timing of "Codex" trending in India on March 22 — three days after the Astral acquisition — is not a coincidence.

How to Start Using Codex

Cloud agent: Go to chatgpt.com/codex. Requires a ChatGPT Plus, Pro, or Team subscription. Connect your GitHub account for repo access. Give it a task in plain English.

CLI: Run npm i -g @openai/codex or brew install --cask codex. Navigate to your project directory. Run the codex command and describe what you want done. It will show you its plan before executing any commands.

API: Use the codex-mini-latest model via the standard OpenAI API. Set up your API key, call the model as you would any other OpenAI model. The 75% caching discount activates automatically for repeated context.

For agentic Python workflows specifically, watch for the Astral integration over the next 6 months — that is when the environment management and linting capabilities will fold into the core product.

Key Takeaways

• Codex is two products: a cloud agent (chatgpt.com/codex) for async background tasks and a CLI tool for local terminal-based coding
• 2 million weekly active users as of March 2026 — up 3x in users and 5x in usage since January
• Astral acquisition (March 19) adds uv, Ruff, and ty to the Codex toolchain — targeting Python environment management and code quality in agentic workflows
• Pricing: included in ChatGPT Plus ($20/month); API access at $1.50/$6.00 per million tokens with 75% caching discount
• Best use case vs competitors: Codex for autonomous batch tasks, Cursor for active real-time coding, Copilot for GitHub-integrated enterprise environments
• India is a key growth market — Python dominance + ChatGPT adoption + Astral acquisition timing drove Codex to trend in India on March 22, 2026