// OpenAI
OpenAI Codex pricing and models
OpenAI’s coding agent, included in every ChatGPT plan including Free, or billed per token with an API key.
All coding toolsPlans checked 2026-10-27
- Free plan
- Yes
- Paid from
- $8/mo
- Own API key
- Yes
- Open source
- No
Plans
Terminal, IDE extension, desktop app and cloud tasks.
| Plan | Price | What you get |
|---|---|---|
| Free | Free | GPT-6 Luna in the desktop app, subject to rollout |
| Go | $8/mo | GPT-6 Luna in the desktop app, subject to rollout |
| Plus | $20/mo | About 15–160 GPT-6.1 Sol messages per five hours; weekly limits may apply |
| Pro | from $100/mo$200 and $500 tiers | No five-hour limit |
| Businessteam | $25/moper user · $20 billed annually | Same five-hour estimates as Plus |
Included models and how they score
LiveBench coding and long-horizon agentic scores, strongest coder first, with what the same heavy month of agent work would cost paying per token.
| Model | Coding | Long-horizon agentic | Same work via API |
|---|---|---|---|
| GPT-6.1 Sol | 96.0 | 68.7 | $1,336/mo |
| GPT-6.1 Terra | 94.3 | 48.6 | $668.00/mo |
| GPT-6 Astra | 93.7 | 77.5 | $6,960/mo |
| GPT-6 Luna | 84.5 | 30.9 | $69.60/mo |
Compare OpenAI Codex with
OpenAI Codex FAQ
How much does OpenAI Codex cost?
OpenAI Codex has a free plan (gpt-6 luna in the desktop app, subject to rollout). Paid plans start at $8/mo (Go). Prices from OpenAI’s pricing page, checked 2026-10-27.
Which models does OpenAI Codex use?
GPT-6.1 Terra, GPT-6.1 Sol, GPT-6 Luna, GPT-6 Sol, GPT-6 Astra.
Can I use OpenAI Codex with my own API key?
Yes — usage is then billed per token by the provider instead of by a plan. A heavy coding-agent month (40K tokens in, 4K out, 20K requests, 70% cached) on GPT-6.1 Sol comes to about $1,336 at list price.
Where these numbers come from
- Plans and prices — each vendor's own pricing page, checked 2026-10-27: OpenAI Codex. Usage allowances change often; check before you buy.
- Model scores — LiveBench coding and long-horizon agentic categories for the models each plan includes, live from the same data as the leaderboard. A tool's own agent loop matters too; these scores rate the model, not the tool.
- “Same work via API” — a heavy coding-agent month (40K tokens in, 4K out, 20K requests, 70% cached input) at OpenRouter list price, for comparing a subscription against paying per token.