01What GPT-6 Sol and Luna Are and How They Work

GPT-6 Sol and Luna are two new AI models from OpenAI that make ChatGPT more powerful and affordable. They are built on the same advances behind Astra-6 but are designed for different tasks. Sol handles complex coding and professional work, while Luna focuses on fast, efficient everyday tasks.

These models are not as powerful as Astra, which remains the best for frontier-level results. Instead, Sol and Luna give you more usage for the same subscription price. They run longer without hitting usage limits and produce clearer, shorter answers with less jargon.

This means you can get more done without upgrading your plan.

02Key Features of GPT-6 Sol and Luna

Both models come with important improvements over earlier GPT versions. Here are the main features:

  • 50% lower API prices – Sol and Luna cost half as much as GPT-5.6, with Sol at $2 per million input tokens and $10 per million output, and Luna at $0.10 and $0.50.
  • Fewer hallucinations – Sol gets facts wrong about half as often as GPT-5.6. Sol Max cuts hallucinations from 92% to 60%, while Luna drops from 93% to 77%.
  • Longer conversations – Better caching for agents and extended context windows mean you can run large projects without running out of tokens.
  • Clearer answers – Responses are shorter, use less confusing jargon, and are more honest about what the AI actually checked.
  • Rolling out today – Available in ChatGPT Work, Codex, and the API for eligible users, with free and Go users getting Luna in the desktop app.

03How to Get Started with GPT-6 Sol and Luna

Access depends on your plan and platform. Follow these steps to start using the models:

  1. Check your plan – Plus, Pro, Business, Enterprise, and Edge users get access in ChatGPT Work and Codex. Free and Go users can try Luna in the desktop app.
  2. Update your app – Ensure you have the latest version of ChatGPT or Codex. The models roll out gradually throughout the day.
  3. Switch to ChatGPT Work – Open the work section of ChatGPT. You’ll see a model selector with options: GPT-6 Sol (with sizes light and medium) and GPT-6 Luna.
  4. Choose your model – For fast, simple tasks, pick Luna. For complex coding or professional work, select Sol medium or light. For frontier tasks, use Astra light instead.
  5. Start using in Hermes Agent – Once available on desktop, you can also use Sol and Luna inside Hermes Agent for agent-based workflows.

04How to Use GPT-6 Sol and Luna in Practice

In this episode, Julian ran a side-by-side test comparing GPT-6 Luna and GPT-6 Sol Medium on a project. Luna produced a simple racing game with basic graphics and limited movement, while Sol created a much nicer-looking game with better design and gameplay.

This shows how to choose the right model. Use Luna for fast, lightweight tasks like quick chatbot agents or simple projects. Use Sol for more demanding work where quality matters, such as coding complex applications or handling professional documents.

For the best results on frontier-level tasks, stick with Astra.

Both models finished the test at about the same time, but Sol’s output was clearly higher quality. The trade-off is that Sol costs more per token than Luna, but still 50% less than GPT-5.6.

05Results and Comparisons from the Episode

Julian tested Sol and Luna against each other and compared them with other models. Here are the key results:

  • Sol vs. Luna quality – Sol produced a much nicer game with better design and gameplay. Luna’s version worked but was very basic.
  • Hallucination reduction – Sol Max dropped from 92% to 60%, Luna from 93% to 77%. That is a significant improvement in factual accuracy.
  • Artificial Analysis ranking – Sol slightly overtakes GPT-5.6 Sol, while Luna is roughly level with GPT-5.6. Opus 5.5 from Anthropic ranks number one overall, followed by Fable 5.1 and GPT-6 Astra Max.
  • Price comparison – Sol API pricing dropped from $4 input / $20 output to $2 / $10. Luna dropped from $0.20 / $1.20 to $0.10 / $0.50.
  • Not all improvements – Some benchmarks like GDP Val AA and AA Briefcase did not improve. The models show a mix of gains and regressions.

06Tips, Limits, and Mistakes to Avoid

To get the most out of GPT-6 Sol and Luna, keep these points in mind:

  • Don’t expect frontier performance – Sol and Luna are not as powerful as Astra. For the best results, use Astra for complex tasks and Sol/Luna for everyday work.
  • Watch the rollout timing – The models roll out gradually. If you don’t see them yet, check for updates and try later in the day.
  • Use Luna for speed, not quality – Luna is fast and cheap, but its output quality is lower. Avoid using it for tasks that need high accuracy or polish.
  • Monitor token usage – Even with longer context windows, large projects can still hit limits. Cache strategically for agent workflows.
  • Check the model selector – In ChatGPT Work, you can switch between Luna, Sol light, Sol medium, and Astra light. Pick the right one for your task to save costs and time.

07Who GPT-6 Sol and Luna Are For and Why They Matter

These models are for anyone using ChatGPT who wants more value from their current subscription. If you hit usage limits often, Sol and Luna let you run longer conversations and more complex projects without upgrading.

Developers and professionals will benefit from Sol’s improved coding and factual accuracy. Everyday users can rely on Luna for fast, simple tasks. The lower API prices also make these models attractive for building AI agents at scale.

Why it matters: OpenAI is competing with Anthropic’s Opus 5.5 and other models. By offering cheaper, more reliable options, Sol and Luna make AI more accessible while pushing the whole industry to improve. You can now do more with less, whether you are a beginner or an expert.