If you follow AI news even loosely, this month has been noisy. OpenAI released GPT-5.6 today — three versions of it, no less — and Anthropic’s Claude Fable 5 came back online last week after a dramatic month (more on that in a separate post). Both are billed as the most capable thing their maker has ever shipped.
So the question I keep getting: which one should we be using?
What actually changed
OpenAI’s release replaces the old jumble of model names with three tiers: Sol is the most powerful, Luna is the fast and cheap one, and Terra sits in the middle for everyday work — roughly the same performance as the previous flagship at half the price. Anthropic’s Fable 5 is a single top-end model, priced above Sol, and pitched squarely at heavyweight work — one early customer reported a code migration that would have taken a team two months being done in a day.
Both are genuinely strong. Both are available today in the tools you may already pay for — ChatGPT and Claude.
The honest comparison
Here’s the part the launch coverage won’t tell you: for the work most owner-led businesses actually do — drafting, summarising, triaging, replying — you would struggle to tell these models apart in a blind test. The differences show up at the extremes: very large software projects, deep research, long multi-step jobs done without supervision.
The gap between the two best models on the market is smaller than the gap between a well-briefed model and a lazily-briefed one. How you use it still beats which one you bought.
Where the choice does matter:
- If your team already lives in one tool, stay put. Switching costs are real and the capability gap doesn’t justify them for everyday work.
- If you’re paying per use through the API, the new mid-tier pricing is worth a look — Terra at half the old price is the quiet story of this launch.
- If you have one genuinely hard job — a big migration, a gnarly analysis — it’s now reasonable to try it on a frontier model before hiring it out. Fable 5 and Sol were built for exactly that.
A small, honest next step
Take one real task your business did this week and run it through both — most of you can do this on existing subscriptions for free. Same brief, both tools, compare the output. Ten minutes, and you’ll have a better answer for your work than any benchmark chart can give you.
And if you can’t tell the difference? That’s not a disappointment. It means you can choose on price and convenience, and get back to work.