OpenAI
APIs and tools for building AI products
Start new thread

GPT-6 Sol & Luna - Frontier AI intelligence, now at half the price

by•
OpenAI expands the GPT-6 family with Sol and Luna with faster, cheaper models trained like Astra. 50% lower API pricing, near-Astra factuality/coding/computer use gains, better prompt caching (90% off cached reads), and alignment improvements. Live in ChatGPT Work, Codex, and via API.

Add a comment

Replies

Best
Hunter
📌

GPT-6 Sol and Luna extend OpenAI's newest model family below Astra, trained the same way but tuned for speed and cost.

What makes it different: Sol beats Claude Opus 5 on Automation Bench at roughly 9% of the cost per task, and matches Claude Fable 5.1 on DeepSWE at about 80% lower cost. Both models also carry over Astra's alignment gains, including fewer misleading claims about their own coding work.

Key features:

  • 50% lower API pricing than GPT-5.6 equivalents

  • Prompt caching improvements, 90% discount on cached input reads

  • Near-Astra factuality, coding, and computer-use performance

Who it's for: developers and teams running high-volume or long-running agent workflows who need frontier-level performance without frontier-level cost.

Try it via , , or through the

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified →

Does the prompt caching for Sol and Luna work the same way as it does for Astra?

Day by day it's getting complete bullshit, useless, previously I was saying everybody that chatgpt is good but it's too bad more and more less than a child memory. useless specially now a days just after astra is launched. chat gpt just moved normal chat gpt veriosn to astra and made normal chatgpt version is bullshit ever no body can expect this. it's not able to change the font weight from a single html and css file. and so on i have seen. you guys are not fixing just launching new version. update should be fixed not to create a completely new version which has so many bugs. The chatgpt says always yes even if something is totally incorrect. is this you guys have Artificial intelligence mind??
Now why don't you reply keep increasing version and instead of fixing in new version it's getting more and more bullshit. you guys are just trying to pull money from pockets not launching the right things.

So far we’re at 5.6 sols, but if the 6-sol version turns out stronger and significantly cheaper... I am totally beyond excited! 🚀

Chat has been such a help with getting all my tasks organized and is a very helpful assistant to invest in. I've explored just a few other agents but none seem to get it the way chat does.

the "9% of the cost per task" number is the headline but for us the thing that actually matters is tokens/sec, not cost per token. we run voice agents where a model that's cheaper but 200ms slower to first token is a worse deal than a pricier one that's fast, since the caller notices the silence way before they'd notice the bill. anyone tried Sol specifically in a low-latency streaming setup yet, or is the benchmarking so far all batch/offline style tasks

Intereting to see GPT 6 Sol and Luna focused on different use cases the lower cost and faster models could be really useful for everyday AI workflows

The 90% discount on cached reads may be the bigger deal here than the headline price cut. Long-running agent work keeps pulling in the same project context, so that cost adds up quietly.