Anthropic released the Claude 5 family this year, and Fable 5 is the first model in it: a new tier that sits above Opus. After running it through our internal evaluation, we've moved our delivery work onto it. This is what that means in practice.
Why we moved
The headline benchmarks matter less to us than the unglamorous parts of engineering, because that's where AI assistance earns its keep on real projects:
- It holds more of a large enterprise codebase in view at once, so changes respect the code that already exists
- It plans multi-step work and verifies its own output as it goes, which means fewer broken intermediate states
- It needs fewer retries to land a change, which compounds into noticeably faster discovery and implementation cycles
None of that shows up in a demo. All of it shows up in a delivery timeline.
How we roll a new model out
We don't switch tools on client work the day something launches. Every model goes through the same sequence:
- It runs on internal work first, where mistakes cost us and nobody else
- We compare its output side by side with the previous model on tasks we already know well
- Only then does it touch client work, with the same review gates as before
Fable 5 cleared that bar faster than any model we have evaluated.
What doesn't change
Senior engineers review every line that ships. The model got better; the guarantee stays the same. Our security posture, our review process, and our instrumentation standards don't move with the model version, and that's the point: the tooling accelerates the work, the experience keeps it safe.
If you want to see what this looks like on your codebase, that's exactly what a workshop is for.