GPT-6 Sol and Luna

GPT-6 Sol and GPT-6 Luna are the two cheaper tiers of OpenAI’s GPT-6 generation, released on 22 September 2026, three weeks after GPT-6 Astra. They replace GPT-5.6 Sol and Luna. There is no GPT-6 Terra, so the OpenAI lineup is now three models: Astra for the hardest problems, Sol for demanding everyday work, and Luna for volume.

What they cost

Sol costs $2 per million input tokens and $10 per million output, half of what GPT-5.6 Sol cost. Luna costs $0.10 and $0.50, down from $0.20 and $1.20. OpenAI says better caching and inference let it pass the savings on. Watch the long-context line: once a prompt passes 272K tokens, input costs double and output costs 1.5 times as much. For comparison, Sol now sits at the same per-token price as Claude Sonnet 5, and Astra costs five times as much as Sol.

Both share Astra’s 1.05M-token context window and 128K output limit.

What you get for it

OpenAI trained Sol and Luna with similar methods to Astra and pitches them as bringing much of what makes Astra good down to prices you can use at scale. On OpenAI’s internal factuality evaluation, Sol makes about half as many mistakes as GPT-5.6 Sol did, approaching Astra’s reliability. Luna is the fast one, built for high-volume work, and at higher effort levels it matches GPT-5.6 Sol at about a hundredth of the cost. OpenAI’s launch benchmarks also put Sol ahead of Claude Opus 5 on some agentic business-workflow tests at a fraction of the cost per task; those comparisons predate Opus 5.5.

GPT-5.6 Luna was already the tier I thought people underrated: a genuinely capable model, and the amount of work you could get through with it on even the cheapest paid plan was remarkable. The new one is cheaper and, by OpenAI’s numbers, better.

Where you’ll find them

At launch, Sol and Luna are in ChatGPT Work (the agent mode that works across your apps and files) and in Codex, for Plus, Pro, Business, Enterprise and Edu accounts, plus the API. Free and Go accounts get Luna in the ChatGPT desktop app. Regular ChatGPT chats don’t offer them yet, so if you only use the chat window, you won’t see them.

How they compare

On capability for the money, these are the strongest models OpenAI has, and on price per token they match or undercut the Claude models they compete with. What decides it for me isn’t the price table. I prefer Claude, so Opus 5.5 stays my default, but that’s a preference.

And for most people, the more useful question isn’t which vendor wins this month. It’s whether you know the one you picked well. Each has its own quirks, and you only feel the difference between them once you really know what you’re doing and use them for hard work. Switching back and forth between ChatGPT and Claude costs you that depth. If you’re on OpenAI’s stack, stay there and learn it properly: Sol is the default, Astra the step up for when Sol falls short.

The Verdict

Best for: Execution work at scale: Sol when the task is hard or mistakes are expensive, Luna for high-volume work where speed and price matter most.