GPT-6 Sol vs Luna: Pricing, Benchmarks and Which to Pick

OpenAI followed GPT-6 Astra with two cheaper models: GPT-6 Sol and GPT-6 Luna. Sol is aimed at complex coding and professional work at a much lower cost than Astra. Luna is built for fast, high-volume everyday tasks. I’ve been running Sol in Codex since launch, and in this guide I’ll cover what both models cost, what OpenAI’s benchmarks show, and how I’d choose between Sol, Luna and Astra.

Prefer to watch? See the full walkthrough on YouTube. I build the whole thing on screen.

What OpenAI released

OpenAI announced GPT-6 Sol and GPT-6 Luna on September 22, 2026, as part of what it calls the GPT-6 universe. According to OpenAI, both were trained with similar methods to Astra and bring its gains in professional work, factuality, coding, computer use and alignment to faster, cheaper models.

• GPT-6 Astra: OpenAI’s most capable model. Choose it when you need the best result and cost matters less.
• GPT-6 Sol: complex coding, agentic workflows and professional work at a fraction of Astra’s price.
• GPT-6 Luna: the most efficient model, for focused, high-volume tasks.

One update since the video: on September 29, OpenAI announced GPT-6.1 Sol. On OpenAI’s pricing page, it lists the same $2 input and $10 output rates as GPT-6 Sol, with cheaper cached input. If you’re starting fresh, check which Sol version your tools default to.

For context on the previous generation, see why ChatGPT 5.6 was available, but not for everyone.

Pricing: 50% cheaper than GPT-5.6

OpenAI cut API prices for both models by 50% compared with the GPT-5.6 promotional pricing. Standard rates per million tokens, for prompts up to 272K input tokens:

• GPT-6 Sol: $2 input, $0.20 cached input, $10 output (GPT-5.6 Sol was $4 / $20).
• GPT-6 Luna: $0.10 input, $0.01 cached input, $0.50 output (GPT-5.6 Luna was $0.20 / $1.20).
• GPT-6 Astra, for comparison: $10 input, $50 output.

So Sol’s list price is about a fifth of Astra’s. Longer prompts above 272K tokens cost more, and Batch and Flex processing cost half. See OpenAI’s pricing page for current rates.

What the benchmarks show

These are OpenAI’s own reported results, so treat them as a starting point and test on your own work:

• AutomationBench (business workflows across apps): GPT-6 Sol at xhigh effort scored 33.2% at $0.27 per task, ahead of Claude Opus 5 at max effort (26.9%) at about 9% of Opus 5’s cost per task.
• Agents’ Last Exam (complex professional workflows): Sol at max effort scored 56.4%, above Opus 5’s highest score in the evaluation, at 60% lower cost per task.
• FrontierCode (merge-ready code changes): Sol improves substantially on GPT-5.6 Sol and matches Claude Fable 5.1 at xhigh effort, at much lower cost.
• DeepSWE v1.1 (real codebases): Sol at max effort scored 68.8%, within 1.1 points of Claude Fable 5’s best, at about 80% lower cost per task. Luna at max scored 66.6%.
• Computer use (OSWorld 2.0): Sol at xhigh scored 60.5%, similar to Opus 5 at medium, at about 80% lower cost. OpenAI says Astra remains its best model for computer use.

OpenAI also reports that Sol makes about half as many factual mistakes as GPT-5.6 Sol on its internal factuality evaluation, and that both models show lower rates of misleading claims about their coding work.

Better prompt caching

OpenAI says it improved prompt caching for GPT-6 to deliver higher cache hit rates by default, with a 90% discount on cached input reads. For agents that reuse a lot of context, that can cut costs meaningfully. You can now change reasoning effort or enable and disable tools without breaking the cache.

In my own Codex sessions, I’ve used only a small share of my weekly limit on Sol at medium effort. That’s my experience, not a guarantee, since usage depends on your plan and tasks. I also use a token-saving tool called RTK alongside it, which I’ve covered in other videos.

If you want to put Sol to work in Codex on real projects, the Codex tutorials in AI Income Lab walk through practical builds step by step.

Where you can use Sol and Luna

• ChatGPT Work and Codex: for Plus, Pro, Business, Enterprise and Edu users.
• Free and Go users: GPT-6 Luna in the desktop app.
• API: gpt-6-sol and gpt-6-luna via the Responses and Chat Completions APIs.
• Not yet in Chat at launch, according to OpenAI’s announcement.

The model pages list a context window of about 1.05 million tokens and reasoning effort from none up to max.

Which model should you use?

• Use Luna for classification, extraction, summaries, simple automations and anything you run thousands of times.
• Use Sol as your default for coding in Codex, agent workflows and most professional work.
• Use Astra for the hardest problems, heavy computer use or when a wrong answer costs more than the tokens.

A practical pattern is to start on Sol at medium effort, raise the effort for hard tasks, and move to Astra only when Sol gets stuck.

Whichever model you choose, cost control matters. See a dashboard that shows where your AI budget is going and how to cut token usage in Claude Code.

Common mistakes to avoid

• Defaulting to Astra for everything. It’s far more expensive and burns through usage quickly.
• Taking benchmarks at face value. These are vendor-reported numbers, so test on your own tasks.
• Ignoring long-prompt pricing. Prompts over 272K tokens are billed at higher rates.
• Breaking your cache. Keep stable context at the start of your prompts so it can be reused.

Want to compare notes on model choice with other builders? Join the AI Income Lab community.

FAQ

How much does GPT-6 Sol cost?

$2 per million input tokens and $10 per million output tokens at standard rates, with cached input at $0.20.

How much does GPT-6 Luna cost?

$0.10 per million input tokens and $0.50 per million output tokens, with cached input at $0.01.

Is GPT-6 Sol better than Claude Opus 5?

On OpenAI’s reported AutomationBench and Agents’ Last Exam results, Sol scored higher at a much lower cost per task. Results on your own work may differ.

Can free users try them?

Free and Go users can access GPT-6 Luna in the ChatGPT desktop app. Sol requires a paid plan or the API.

Conclusion

GPT-6 Sol and Luna make frontier-level work much cheaper. Sol is a strong default for coding and agent work, Luna handles volume, and Astra stays in reserve for the hardest jobs. Watch the video above to see the benchmarks and Sol running in Codex.