OpenAI Codex Pricing Explained: Plans, Usage Limits and Hidden Costs in 2026
OpenAI Codex starts free, with paid access beginning at $8/month (Go), $20/month (Plus), $100–$500/month (Pro), and $20/user/month for Business (billed annually, 2+ users) — with API-key usage billed separately per token. The main hidden cost is usage limits: message allowances vary widely by model (5–45 messages per 5 hours on GPT-6 Astra vs 350–3,000 on GPT-6 Luna for Plus), and Fast/Ultrafast modes consume your allowance 2.5x–8x faster.
TL;DR
- Codex pricing tiers: Free ($0), Go ($8/month), Plus ($20/month), Pro (from $100/month at $100/$200/$500 tiers), Business ($20/user/month billed annually, $25 monthly, 2+ users), and Enterprise (contact sales).
- Business Premium seats cost $100/user/month annually ($125 monthly) and include 5x more usage than Standard with no 5-hour limit.
- Usage limits are estimates, not fixed caps: Plus and Standard Business get roughly 5–45 messages per 5 hours on GPT-6 Astra, 15–160 on GPT-6.1 Sol, and 350–3,000 on GPT-6 Luna.
- Fast mode consumes included usage at 2.5x the Standard rate; Ultrafast modes consume at 8x — a major hidden cost multiplier.
- API-key usage is billed separately per token (e.g. GPT-6.1 Sol at $2.00 input / $10.00 output per 1M tokens) and cannot be estimated from subscription allowances.
- Codex usage is shared with ChatGPT Work and other agentic features like ChatGPT for Excel on Plus and Pro.
- The Codex desktop app was macOS-only as of February 2026, with no Windows or Linux support.
- Pro plans currently have no five-hour limit; Enterprise/Edu with flexible pricing has no fixed rate limits and scales with credits.
Head to head
| Dimension | OpenAI Codex |
|---|---|
| Entry price | $0 (Free); $8/mo Go; $20/mo Plus 🏆 |
| Top personal tier | Pro at $100, $200, or $500/month; Ultrafast mode on $500 tier 🏆 |
| Business pricing | $20/user/mo Standard (annual), $25 monthly; $100/user/mo Premium (annual), $125 monthly; 2+ users 🏆 |
| Usage limits | Estimates only: 5-45 msgs/5h (GPT-6 Astra) to 350-3,000 (GPT-6 Luna) on Plus and Standard Business 🏆 |
| Hidden cost factors | Fast mode 2.5x usage; Ultrafast 8x; API billed per token separately; usage shared with ChatGPT Work and ChatGPT for Excel 🏆 |
| Platform availability | Web, CLI, IDE extension, iOS; desktop app macOS-only (as of Feb 2026) 🏆 |
| Free tier | Yes — $0/month with GPT-6 Luna at Standard speed in the desktop app 🏆 |
| API option | API key: CLI/SDK/IDE only, no cloud features, billed per token (e.g. GPT-6.1 Sol $2.00 in / $10.00 out per 1M) 🏆 |
The contenders
OpenAI Codex is an AI coding agent available on the web, in the CLI, in an IDE extension, on iOS, and in a macOS desktop app launched February 2, 2026. Its headline differentiator is a multi-agent architecture: multiple AI coding agents run in parallel on isolated Git worktrees, handling tasks up to 30 minutes each without merge conflicts. It reached 1 million+ active developers with 20x usage growth since August 2025. Codex is bundled with ChatGPT plans rather than sold standalone — Free, Go, Plus, Pro, Business, and Enterprise tiers all include it, with usage shared across ChatGPT Work and other agentic features. Reviewers report success rates of 85–90% on well-scoped maintenance work by March 2026, with strong error handling, multi-turn refinement, and a preview system generating 2–4 implementation approaches per task. Notable limitations include macOS-only desktop availability, no user control over model selection, and the need for human code review on security and quality.
pricing 0/month
- + Free tier and a low $8/month Go plan make it cheap to evaluate
- + Bundled with ChatGPT plans, so Plus/Pro subscribers get Codex at no extra cost
- + Multi-agent parallel development with worktrees eliminates merge conflicts between concurrent agents
- + Reported 85-90% success rate on well-scoped maintenance tasks, up from 40-60% in 2025
- + Business plans include enterprise-grade security defaults: SSO, MFA, no training on business data
- + Mix-and-match Standard and Premium Business seat types to fit different usage levels per employee
- − Usage limits are estimates that vary by model, task size, and context — prompt length alone isn't a reliable cost predictor
- − Fast and Ultrafast modes burn included usage 2.5x-8x faster, a hidden multiplier on effective cost
- − Desktop app is macOS-only as of February 2026, with no Windows or Linux support
- − No user control over model selection — Codex picks internally based on task complexity
- − API token prices are separate from subscription usage and can't be used to estimate included tasks
- − Cloud features (GitHub code review, Slack) are unavailable on API-key plans
Verdict
Winner: none
This is a single-product pricing explainer, not a head-to-head comparison. The verdict instead: Codex is affordable to start (free tier, $8 Go plan) but costs scale unpredictably because usage limits are model- and speed-mode-dependent rather than fixed.
| Option | Overall | Breakdown |
|---|---|---|
| OpenAI Codex | 8 | : 8: 6: 8: 5 |
Pricing compared: every Codex plan and what you get
The hidden costs: usage limits, speed multipliers, and shared allowances
Is Codex worth it for a small business in 2026?
Which should you pick?
- :
- :
- :
- :
- :
- :
FAQ
How much does OpenAI Codex cost in 2026?
Codex is bundled with ChatGPT plans: Free at $0/month, Go at $8/month, Plus at $20/month, Pro at $100/$200/$500 per month, Business at $20/user/month billed annually ($25 monthly, 2+ users), and Enterprise by contacting sales. Business Premium seats cost $100/user/month annually ($125 monthly) with 5x more usage and no 5-hour limit.
What are Codex usage limits, and are they fixed?
No — OpenAI describes them as estimates, not fixed message limits. Plus and Standard Business users get roughly 5–45 local messages per five-hour period on GPT-6 Astra, 15–160 on GPT-6.1 Sol, and 350–3,000 on GPT-6 Luna. Cloud tasks consume more allowance than local messages, weekly limits may apply, and Pro plans currently have no five-hour limit.
What are the hidden costs of Codex?
Three main ones: Fast mode consumes included usage at 2.5x the Standard rate and Ultrafast at 8x; your allowance is shared with ChatGPT Work and other agentic features like ChatGPT for Excel; and API-key usage is billed separately per token, so it can't be estimated from subscription allowances. Regional processing endpoints also add a 10% uplift for models released on or after March 5, 2026.
Can I extend my Codex usage if I hit the limit?
Yes — Plus, Pro, and Business plans can flexibly extend usage with ChatGPT credits. Credits are consumed at different rates depending on the speed mode (2x for Fast, 6x for Ultrafast relative to Standard). Check the usage dashboard for current limits and reset times.
Is Codex available on Windows or Linux?
As of February 2026, the Codex desktop app is macOS-only. However, Codex is also available on the web, in the CLI, in the IDE extension, and on iOS, which work across platforms. Windows and Linux users wanting a dedicated desktop app may need to consider alternatives like Claude Code or Cursor.
Does the Business plan train on my company's data?
No. OpenAI states there is no training on your business data by default on the Business plan, which also includes a secure dedicated workspace, SAML SSO, MFA, centralized billing, usage analytics, and spend controls. Enterprise adds SCIM, EKM, RBAC, audit logs, and data-residency controls.
Should I use the subscription or the API for automation?
The API key option is built for automation in shared environments like CI: you get Codex in the CLI, SDK, or IDE extension and pay per token based on API pricing. The trade-off is that cloud-based features like GitHub code review and Slack integration are unavailable, and model availability follows whatever models your key can access.