Is Codex now bigger than Cursor and Claude Code?

By active weekly users, yes. OpenAI announced 5 million weekly Codex users on June 2, 2026, including the new Sites, Annotations, and six role-specific plugins. Cursor reports roughly $3B annualized revenue and millions of users but hasn't published a precise weekly active number. Claude Code has had explosive growth since the June 15, 2026 billing change but remains smaller than Codex on raw WAU. On engineering productivity per dev, however, Claude Code and Cursor still rank higher in developer-satisfaction surveys.

What's the best AI coding tool to use in June 2026?

It depends on workflow. For terminal-driven autonomous coding: Claude Code. For IDE flow with the smoothest in-editor experience: Cursor 3. For ChatGPT-integrated work that includes sites, knowledge work, and code: Codex. Many shipping teams now run all three — Claude Code for autonomous backend work, Cursor for UI work, and Codex CLI for ChatGPT-tied workflows. Cost: $20–$200/month per tool, depending on plan.

What changed with Codex on June 2, 2026?

OpenAI's 'Intelligence at Work' update added: (1) Codex Sites — hosted, interactive sites from prompts; (2) Annotations — scoped, in-place edits to existing files and sites; (3) six role-specific plugins covering 62 business apps and 110 skills for analysts, designers, investors, bankers, and more; (4) AWS Bedrock GA for Codex; (5) the 5M weekly active user milestone. About 20% of weekly users are now non-engineers — knowledge workers.

Is Cursor still worth $20/month if Codex is the volume leader?

Yes, especially for IDE-first developers. Cursor 3 still leads on inline edit quality, predictive autocomplete (Cursor Tab), and the smoothest 'one window' editing experience. Codex is winning on knowledge-worker workflows and ChatGPT integration; Cursor is winning on traditional developer ergonomics. Most professional devs still pay for Cursor; the 5M Codex number is heavily inflated by non-developers using Annotations and Sites.

Quick Answer

Codex 5M vs Cursor vs Claude Code: AI Coding June 2026

Published: June 8, 2026

Codex 5M vs Cursor 3 vs Claude Code: State of AI Coding June 2026

OpenAI announced 5 million weekly Codex users on June 2, 2026 — a number that reframes the AI coding race. Cursor still dominates IDE flow. Claude Code dominates autonomous terminal workflows. Codex now dominates raw scale. Here’s the honest state of the AI coding stack in June 2026.

Last verified: June 8, 2026

TL;DR

Tool	Weekly users	Best for	Starts at
OpenAI Codex	5M+	Knowledge work + code + ChatGPT integration	$25/user/mo
Cursor 3	”Millions” (≈$3B ARR)	IDE flow, inline edits, autocomplete	$20/mo Pro
Claude Code	Smaller but growing fast	Autonomous terminal coding	$20/mo Pro
Devin Desktop (was Windsurf)	Smaller	Multi-agent IDE	$20/mo
GitHub Copilot	Largest installed base	Inline completion at enterprises	$19/seat/mo

What each one actually is in June 2026

OpenAI Codex

The June 2 update made Codex a workspace, not a CLI:

Codex CLI — terminal agent for repos
Codex Sites — hosted sites from prompts
Annotations — scoped, in-place edits across files
Six role plugins — analyst, designer, investor, banker, marketer, ops
5M weekly users — 20% non-engineers
AWS Bedrock GA, ChatGPT Business included

Cursor 3

Cursor is the IDE-first king:

Cursor 3 with Composer + Opus 4.8 backend
Inline edit, Cursor Tab autocomplete, Composer multi-file
Cursor 3.6 added Auto Review + Bugbot
~$3B annualized revenue, valuation ~$60B
$20 Pro, $40 Ultra; usage-based on top

Claude Code

Terminal-native autonomous:

Anthropic’s flagship coding tool
Claude Opus 4.8 + Dynamic Workflows (up to 1000 subagents)
June 15, 2026 billing change adds metered credits on top of Max plans
Best autonomous “describe and walk away” loop
$20 Pro, $100/$200 Max tiers

Devin Desktop (was Windsurf, rebrand June 2, 2026)

Cognition rebranded Windsurf as Devin Desktop on the same day as the Codex update
Agent Command Center as default surface
Supports Agent Client Protocol (ACP): Codex, Claude Code, and others slot in
$20/mo

GitHub Copilot

Enterprise-default. Largest install base via GitHub. $19/seat/mo. Strongest IP indemnity story.

Where each one wins (real-world)

Workflow	Best tool
Inline edits, codebase navigation, polish	Cursor 3
”Build this feature from scratch, ping me when done”	Claude Code
Knowledge work + code in one app	Codex
Marketing site, internal tool	Codex Sites or Lovable
Multi-agent parallel work	Devin Desktop or Grok Build
Enterprise IP / compliance	GitHub Copilot
Background agent on a real repo	Claude Code + Codex CLI

What 5M weekly users actually means

The headline is misleading. Of those 5M:

~80% are engineers using Codex CLI, Annotations, or VS Code Codex
~20% are knowledge workers using Sites and role plugins (analyst, banker, etc.)

So Codex’s coding WAU is ~4M. That still likely makes it the largest AI coding product by raw weekly use, but the gap to Cursor is smaller than the 5M number implies — and Cursor users tend to be deeper, more daily, more revenue-rich.

Pricing in June 2026

Tool	Entry	Power user	What you get
Cursor 3	$20/mo Pro	$40 Ultra + usage	Best IDE flow
Claude Code	$20 Pro	$200 Max 20x + metered (Jun 15)	Best autonomy
Codex	$25/user (Business)	$60/user (Codex Pro)	Best workspace integration
Devin Desktop	$20/mo	$60 team	Multi-agent ACP
GitHub Copilot	$19/seat	$39 Enterprise	IP indemnity
Augment Free	$0	$0	Free repo context
Gemini CLI Free	$0	$0	Free high-volume

The composable stack pros are actually running

The June 2026 reality is most serious teams run multiple tools:

Cursor 3  →  for fast inline edits + UI work
Claude Code  →  for autonomous "do this whole feature" tasks
Codex CLI  →  for ChatGPT-tied workflows + Sites
Augment (free)  →  for codebase-wide context
Gemini CLI (free)  →  for high-volume bulk tasks
GitHub Copilot  →  for enterprise IP indemnity

Total cost for a serious solo dev: roughly $40–$80/month.

The four big shifts in June 2026

Codex Sites is the first credible competitor to Lovable/Bolt/v0 with a 5M-user funnel feeding it
Windsurf → Devin Desktop rebrand consolidates Cognition’s narrative; ACP makes the IDE multi-agent
Claude Code’s June 15 billing change adds metered credit packs on top of Max — Anthropic is signaling that demand is exceeding plan caps
Knowledge workers are now 20% of Codex users — AI coding is becoming AI working

Who should switch?

If you…	Do this
Currently use Cursor and ship daily	Stay. Add Claude Code for big tasks.
Currently use Claude Code and prefer terminal	Stay. Try Codex CLI for ChatGPT workflows.
Already pay for ChatGPT Business	Try Codex Sites + Annotations.
Are not a developer but want to build apps	Lovable still wins. Codex Sites is #2.
Are at a regulated enterprise	GitHub Copilot is still the safest default.

Bottom line

The AI coding market is no longer a one-winner race. Cursor wins on flow, Claude Code wins on autonomy, Codex wins on scale and knowledge-worker reach. The smartest move in June 2026 is to run two or three of them in parallel and pay $50–$80/month total — far less than one engineer-hour of time saved per week.

The interesting story isn’t “who’s winning.” It’s that coding agents and knowledge-work agents are merging into the same tool — and Codex got there first.