Guide
Which Codex model should you use?
Astra, Sol or Luna: what each model of Codex, OpenAI’s coding agent, is for, what it costs, and how to last until the reset.
At a glance
| Model | Best for | API price | Relative cost |
|---|---|---|---|
| GPT-6 Astra | The most capable: complex end-to-end work across code, apps and research that needs sustained reasoning. | $10 / $50 | ×5 |
| GPT-6.1 Sol | Near-Astra performance on complex work, at a much lower cost. The newest. | $2 / $10 | ×1 |
| GPT-6 Sol | The everyday model for complex coding and agent tasks. Replaces GPT-5.5 on paid plans. | $2 / $10 | ×1 |
| GPT-6 Luna | The most efficient: high volumes of focused tasks, summarising, extraction, focused coding. Replaces GPT-5.5 on Free and Go. | $0.1 / $0.5 | ×0.05 |
API price in US dollars per million tokens, input / output, standard context. Relative cost: price ratio, GPT-6 Sol = 1. On a ChatGPT plan you don’t pay these prices: they give an order of magnitude.
What about GPT-5.5?
GPT-5.5 is a legacy model, retired on October 14, 2026. OpenAI replaces it with GPT-6 Sol on paid plans and GPT-6 Luna on Free and Go.
What about your ChatGPT plan?
Codex is included with ChatGPT Plus, Pro, Go and Business, with its own limits: a 5-hour session and a weekly limit (a 30-day window on Go).
OpenAI advises starting from your account’s default setting and raising the reasoning effort when a task needs more planning.
Like Anthropic, OpenAI doesn’t publish the exact weight of each model on ChatGPT limits. The relative cost above remains the best public order of magnitude.
Using Claude too? Read the Claude model guide
What JaugeAI does for you
- Your Codex limits live, 5-hour session and week, next to Claude’s.
- The forecast: when the limit will be full at your pace.
- What your activity is worth at API prices, by project, from Codex’s local logs.
Free on Mac. On the App Store for iPhone and Apple Watch.
All guides: limits, resets, usage credits