Anthropic dropped Opus 5 on Friday.
It’s the new heavyweight. But here is the twist. It is smaller than Fable 5. It is also cheaper. And, crucially, it has fewer restrictions. You might think smaller means worse. The data says otherwise.
Opus 5 actually beats Fable 5 on several benchmarks listed in the launch announcement. It isn’t just a step down in performance, it’s a shift in philosophy.
Why Opus 5 Outperforms Fable 5 Despite Being Smaller
Anthropic released Opus 5 just two months after Opus 4.6 hit the streets on May 28. The timeline is tight. Mythos 5, Fable 6, and Sonnet 5 all came out in June. Haiku is the only one still waiting for a version 5 update.
So why switch?
In the launch post, Anthropic highlighted a specific strength. Opus 5 is much stronger at verifying its own work. It iterates carefully until it succeeds. This isn’t just marketing fluff. The benchmarks showed Opus 5 writing its own computer vision pipeline from an incomplete prompt. It fixed the gaps. It got it right.
“Opus 5 was much stronger at verifying its work and iterating carefully.”
This self-correction capability makes it robust for complex tasks where Fable 5 might hallucinate or give up.
Opus 5 Restrictions: How It Compares to Fable 5
Privacy advocates were nervous about Fable 5. Why? Because of the 30-day data retention policy. It covered Fable and Mythos. It raised red flags for users who care about their data staying local.
Opus 5 fixes that.
It is free from those retention policies. Your data stays yours. This is a significant win for privacy-conscious developers.
There are still guards up, obviously. Cybersecurity tasks remain tricky. Opus 5 will not help you scan for vulnerabilities in software binaries. That is blocked. But you can search for vulnerabilities in source code. Anthropic argues the latter is more likely used for defense.
The classifiers engage 85% less often with Opus 5 compared to Fable 5. A lighter touch. A smoother ride.
Is Opus 5 the Better Choice Over Fable 5?
Let’s look at the trade-offs.
- Cost : Opus 5 is cheaper.
- Freedom : Fewer restrictive classifiers.
- Performance : Beats Fable 5 on key benchmarks.
- Privacy : No 30-day data retention.
Fable 5 is the bigger model. It might still win on raw reasoning power in extremely niche, heavy-lift scenarios. But for most use cases? Opus 5 wins.
It’s faster. It’s less annoying to work with. It verifies itself.
Anthropic isn’t just releasing models anymore. They are releasing workflows. Opus 5 feels less like a brute-force hammer and more like a precision tool.
Which one will you use?
Probably the one that doesn’t lock your data away for a month.
