OpenAI's own rule: "Choose a model based on the reasoning your task requires, latency, and cost." Here's how that plays out across the GPT-6 family.
Prices (per 1M tokens, API)
| Model | Input | Cached input | Output |
|---|---|---|---|
| gpt-6-luna | $0.10 | $0.01 | $0.50 |
| gpt-6-sol | $2.00 | $0.20 | $10.00 |
| gpt-6-astra | $10.00 | $1.00 | $50.00 |
- Luna is 20x cheaper than Sol on both input and output.
- Astra is 5x the price of Sol.
- Cached input is 90% cheaper than fresh input on every model.
- Prompts over 272K input tokens cost more (Sol: $4 input / $15 output).
Which model for which job
Luna: efficient, repeatable work at scale
- Summarizing documents
- Extracting information
- Answering quick questions
- Any high-volume job with a clear goal
Sol: strong reasoning on demanding tasks
- Building features
- Reviewing code
- Debugging
- Analyzing data
Astra: your top tier
- The most complex, multi-part projects
- Harder science and maths problems
The four-step test
- Write down what "good" looks like for the task: the fields you need, the format, a few sample inputs.
- Run it on Luna first. Check the output against your samples.
- Passes? Keep Luna. You're paying a twentieth of Sol's price.
- Fails? Move up to Sol, then Astra only if Sol still misses.
Speed and cost tips
- Sol and Luna support
reasoning_effort: "none". Astra doesn't; uselowthere instead. - In Chat Completions, Sol and Luna only support function calling with
reasoning_effort: "none". - Reusing the same context across calls? Cached input is 90% off.
Where you can use them
- API: all three models.
- ChatGPT Work and Codex: Sol and Luna for Plus, Pro, Business, Enterprise and Edu.
- Free and Go users: can try Luna in the desktop app.
- GitHub Copilot: Luna on Pro and above; Sol on Pro+, Max, Business and Enterprise.
Sources: OpenAI developer community announcement, OpenAI API model guidance, VentureBeat, TechCrunch, GitHub Changelog, aipricing.guru.