OpenAI just launched GPT-6 Sol and Luna, two new models that sit below the flagship GPT-6 Astra and cut API prices roughly in half. If you’re deciding whether to switch, here’s everything that changed, pricing, benchmarks, and access, in one place.
What Are GPT-6 Sol and Luna?
GPT-6 Sol and Luna are the newest additions to OpenAI’s GPT-6 lineup, released on September 22, 2026, joining the GPT-6 Astra flagship model that launched earlier the same month. Both models were trained using methods similar to Astra’s, with the goal of bringing Astra’s improvements to faster and more affordable versions. marktechpost
The two models are built for different jobs. Luna is designed for fast responses at high volume, while Sol carries more reasoning capability, and both models outperform their predecessors on OpenAI’s internal benchmark tests. In plain terms: Sol is the one you reach for on complex coding or multi-step agentic work, and Luna is the one you reach for when you need to run thousands of cheap, quick tasks, summarization, classification, routine chat. MacRumors
How They Fit Into the GPT-6 Family
Q: Where do Sol and Luna sit relative to GPT-6 Astra?
A: They’re the mid- and lower-tier models beneath Astra. The GPT-6 family now has three tiers, Astra is the top model reserved for the hardest work, Sol targets complex coding and professional tasks at a lower cost, and Luna handles high-frequency, lower-complexity jobs. Notably, as of this release there is no GPT-6 Terra model, despite that name having circulated in earlier speculation. marktechpostthenewstack
Both GPT-6 Sol and Luna are API-only releases. They are live in the OpenAI API under the names gpt-6-sol and gpt-6-luna, and since they are proprietary, closed models, there are no downloadable weights to self-host. marktechpost
GPT-6 Sol and Luna Pricing: How Much Cheaper Are They?
The headline change with GPT-6 Sol and Luna is cost. Prices dropped 50% compared to the equivalent GPT-5.6 models, Sol is priced at $2 per million input tokens and $10 per million output tokens, while Luna comes in at $0.10 input and $0.50 output per million tokens. OpenAI framed this explicitly as a strategic move: the company said its improvements in caching and inference let it serve these models more cheaply, and it’s passing those savings on to users by cutting API prices for Sol and Luna by half relative to GPT-5.6 promotional pricing. x9to5google
Here’s how the new pricing compares:
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Best For |
| GPT-6 Sol | $2.00 | $10.00 | Complex coding, agentic workflows, professional tasks |
| GPT-6 Luna | $0.10 | $0.50 | High-volume, high-frequency, lower-complexity jobs |
| GPT-5.6 (prior gen, comparable tier) | ~2x current Sol/Luna pricing | ~2x current Sol/Luna pricing | Superseded by GPT-6 Sol/Luna |
This pricing structure is a big part of why GPT-6 Sol and Luna are being discussed as a cost play rather than purely a capability play, OpenAI is explicitly competing on the price-to-intelligence ratio, not just raw benchmark scores.
GPT-6 Sol and Luna Benchmarks and Performance
Q: Are GPT-6 Sol and Luna actually better than their predecessors, or just cheaper?
A: Both, according to OpenAI’s own testing, at least. GPT-6 Sol makes about half as many mistakes as GPT-5.6 Sol, and both new models beat their predecessors across OpenAI’s benchmark suite. MacRumors
On efficiency specifically, one industry report noted that Sol running at its highest reasoning setting beats Opus 5 Max on AutomationBench while costing only about 9% as much, and that Luna scores 66.6% on the DeepSWE v1.1 benchmark while costing roughly 93% less per task than Opus 5. marktechpost
Alignment work also got attention in this release. Coverage pointed out that OpenAI is emphasizing its safety and alignment work for this release more than usual, following a recent industry incident, and OpenAI reports that both new models improve on their GPT-5.6 counterparts across internal alignment evaluations, including producing fewer misleading outputs. thenewstack
Where Can You Access GPT-6 Sol and Luna?
Availability is staggered by product and account tier. As of launch:
- API access: Both models are live now as gpt-6-sol and gpt-6-luna.
- ChatGPT Work and Codex: Rolling out today for Plus, Pro, Business, Enterprise, and Edu users. MacRumors
- Free and Go tiers: GPT-6 Luna is available to Free and Go users, but only in the desktop app. MacRumors
- Standard ChatGPT chat mode: Not yet available, both models are currently limited to Work and Codex surfaces. MacRumors
- Cloud platforms: Also listed as available on Amazon Bedrock for enterprise deployment.
If you’re on a free ChatGPT plan and want to try GPT-6 Sol and Luna, Luna in the desktop app is currently your only entry point, Sol requires a paid tier.
GPT-6 Sol and Luna vs Claude Opus 5.5: How Do They Compare?
The timing of this release wasn’t a coincidence in the eyes of most coverage. Anthropic and OpenAI both launched new models within roughly an hour of each other on the same day, Claude Opus 5.5 from Anthropic, followed shortly after by GPT-6 Sol and Luna from OpenAI. Anthropic claimed Opus 5.5 beat GPT-6 Astra across most measured categories, including agentic coding and knowledge work, despite Opus 5.5 also running at lower compute cost. 9to5google9to5google
One analysis summed up the competitive dynamic plainly: OpenAI cut GPT-6 pricing by half specifically to compete with Anthropic on cost, but Opus 5.5’s own launch had already reset the comparison, and as of this writing no one had run the two head-to-head yet. thenewstack
For anyone choosing between the two ecosystems, the practical takeaway is that GPT-6 Sol and Luna compete primarily on price-efficiency, while Anthropic’s Opus 5.5 is being positioned on raw capability at a similarly reduced cost, meaning the real deciding factor right now is task fit rather than a clear universal winner.
Frequently Asked Questions
Q: When did GPT-6 Sol and Luna launch?
A: September 22, 2026, shortly after GPT-6 Astra and within about an hour of Anthropic’s Claude Opus 5.5 announcement.
Q: Is GPT-6 Sol or Luna better for coding?
A: Sol. It carries more reasoning capability and is positioned for complex coding and professional workflows, while Luna is tuned for fast, high-volume, lower-complexity tasks.
Q: Can I use GPT-6 Sol and Luna for free?
A: Only Luna, and only through the ChatGPT desktop app on Free and Go tiers. Sol requires a paid ChatGPT plan or API access.
Q: Is there a GPT-6 Terra model?
A: No. Despite earlier speculation, OpenAI released only Sol and Luna alongside Astra, Terra does not exist as part of this lineup.
Q: How much cheaper are GPT-6 Sol and Luna than GPT-5.6?
A: Roughly 50% cheaper per million tokens, on both input and output pricing, across both models.
Conclusion
GPT-6 Sol and Luna mark OpenAI’s clearest move yet toward competing on cost rather than just raw capability, half-price tokens, benchmark gains over GPT-5.6, and a staggered rollout that currently favors paid and desktop users. Whether they hold up against Claude Opus 5.5 in real-world use is still an open question nobody has formally tested yet, but for now, GPT-6 Sol and Luna give developers a meaningfully cheaper way to access GPT-6-era capability.