OpenAI just announced ChatGPT 5.6, and its new flagship, Sol, is the closest any model has come to Anthropic’s frontier. For serious, high-stakes work, though, Claude is still the model I would reach for right now. OpenAI is winning on cost, and Anthropic is still winning on raw capability.
At TJ Digital, we run roughly 90% of our daily work in Claude and test every major model as it ships, so this comparison comes from using these models every day. On OpenAI’s own Terminal-Bench 2.1 numbers, Sol scored 88.8% against Claude Mythos 5 at 88.0%. That is close enough that neither one has a clear lead.
One quick note on the names. Claude Fable 5 and Mythos 5 are Anthropic’s frontier tier, which sits above Opus. They run on the same underlying model, with Fable carrying extra safeguards for sensitive work.
What are Sol, Terra, and Luna?
OpenAI released three models under the 5.6 name, and none of them are publicly available yet. We are going off what OpenAI has shared so far.
- Sol is the flagship, built for the hardest coding, research, and reasoning tasks.
- Terra is the everyday model. OpenAI says it matches ChatGPT 5.5 at about half the cost.
- Luna is the fast, cheap option for high-volume work where speed matters more than depth.
Here is how the pricing lines up against Anthropic’s models, using each company’s published rates per million tokens.
| Model | Input | Output | Best suited for |
| ChatGPT 5.6 Sol | $5 | $30 | OpenAI’s new flagship for complex work |
| ChatGPT 5.6 Terra | $2.50 | $15 | Everyday work at roughly half of 5.5’s cost |
| ChatGPT 5.6 Luna | $1 | $6 | High-volume, cost-sensitive tasks |
| Claude Fable 5 | $10 | $50 | Anthropic’s frontier tier for the hardest work |
| Claude Opus 4.8 | $5 | $25 | Anthropic’s widely available flagship |
Why ChatGPT 5.6 is cheaper than Claude
This is the one area where I think OpenAI is legitimately ahead. Anthropic keeps shipping smarter models, but almost always at the same or higher cost than the model before it. OpenAI is going the other direction and attacking price directly.
Everyone serious about AI wants to use Anthropic’s models. The problem nobody wants to say out loud is that they are getting harder and harder to afford. If Terra really is as good as ChatGPT 5.5 at half the cost, a lot of people are going to switch.
In my view, 5.5 is already about as capable as Claude Opus 4.8. I covered where 5.5 lands in our ChatGPT 5.5 breakdown, and a cheaper model at that level changes the math for a lot of businesses.
Does ChatGPT 5.6 Sol really match Claude’s frontier models?
OpenAI clearly wants Sol to be seen as comparable to Mythos. So far, I am not convinced, and there are two reasons why.
First, OpenAI has only shared a handful of benchmarks. Model makers always pick the tests where their model looks best, so a short list of wins tells you very little.
Second, on some of those benchmarks they do not even show where Mythos lands. When they do, they compare against Mythos Preview, a model that is already more than two months old. You can check that against OpenAI’s announcement.
ChatGPT 5.5 is already a strong model, and 5.6 Sol looks like a real jump up from it. But if Sol were truly on par with Fable or Mythos 5, OpenAI would have a much easier time proving it.
Why OpenAI is talking about the government
OpenAI spent a good chunk of its announcement on how it is cooperating with the United States government to reduce cybersecurity and biological weapon risks. Access is starting as a limited preview to a small group of vetted partners while a federal framework gets worked out.
Part of this looks like the new normal for frontier releases, at least until there is a better way to verify the safety of these models. Part of it also felt like OpenAI wanting to show that Anthropic is not the only lab with a model big enough to worry about.
For now, Anthropic still holds a real lead on the most capable frontier models. You can see that pace in Anthropic’s recent Claude updates. Anything can happen, but I do not see anyone catching them this year.
Are AI subscriptions actually a good deal?
Here is the part that matters most for the average business owner. If you are paying for a ChatGPT or Claude subscription, you are getting a heavily subsidized rate. You are only paying somewhere between 4% and 12% of the actual usage cost.
These flat monthly plans are funded by investor money while these companies build market share. That is why heavy users hit rate limits and why frontier models still cost so much per token on the API. It also means the price gap between Sol and Claude matters far less inside a normal subscription than it does on raw API pricing.
So if the work you need to do fits inside one of these subscriptions, I still think Claude is the best choice right now.
Should you switch from Claude to ChatGPT 5.6?
My answer comes down to what you are trying to do.
For big, important, ambitious projects where you need the smartest model available, use Claude. It is still ahead on the hardest reasoning and long-context work, which is exactly where those projects live. It also looks like we will have Fable 5 back any day now.
For high-volume, cost-sensitive work, ChatGPT 5.6 Terra and Luna are worth a serious look once they are public. The smart move for most teams is to run critical work on Claude and test the cheaper OpenAI tiers on tasks where top-end quality is not the point. If you want a fuller map of which model does what, we keep that updated as new models ship.
Common questions about ChatGPT 5.6 and Claude
Is ChatGPT 5.6 available to the public yet?
No. Sol, Terra, and Luna launched as a limited preview for a small group of approved partners through OpenAI’s API and Codex. OpenAI says broader access is coming in the following weeks.
How much does ChatGPT 5.6 cost compared to Claude?
Sol is $5 per million input tokens and $30 per million output tokens, which matches ChatGPT 5.5. Claude Fable 5 runs about $10 input and $50 output, so Sol is roughly half the price of Anthropic’s frontier tier.
When is Claude Fable 5 coming back?
Anthropic’s most capable public models have been temporarily restricted, and it looks like Fable 5 will be available again very soon. I would expect it back within a few days.
Is Sol or Claude better for coding?
On OpenAI’s Terminal-Bench 2.1 results, Sol and Claude Mythos 5 sit nearly tied at 88.8% and 88.0%. For most real coding work, the two trade the lead by small margins, so access and cost usually decide it.
How to stay ready as AI models change
Whatever model leads next quarter, the businesses that win are the ones with a system built to move between models without missing a step. At TJ Digital, we build that system for you, so your marketing keeps producing no matter which lab is on top this month. Reach out to me and we can map out what that looks like for your business.