Either I’ve already beaten the fallback mechanisms in Fable 5 or they don’t use Opus 4.8 / Sonnet 4.6 / Haiku 4.5 as the fallback. I’ve run several different tasks about upgrading our training library through each model and the output from Fable looks nothing like the others.
I haven’t read the Fable 5 output in detail because I’m on mobile and can’t test it right now, but even if it’s sandbagging it’s still much better than the other model options.
Either I’ve already beaten the fallback mechanisms in Fable 5 or they don’t use Opus 4.8 / Sonnet 4.6 / Haiku 4.5 as the fallback. I’ve run several different tasks about upgrading our training library through each model and the output from Fable looks nothing like the others.
Unless it has introduced subtle bugs I guess? But it was much faster to identify the key changes necessary and provided the information in a better explained and clearer way. And the code looks better designed at a glance.
I haven’t read the Fable 5 output in detail because I’m on mobile and can’t test it right now, but even if it’s sandbagging it’s still much better than the other model options.
Demo with modifying our training library to add a recently developed parallelism technique:
Fable 5: https://claude.ai/share/17f83ad7-0e25-4d47-956c-ad5aa84a2af7
Opus 4.8: https://claude.ai/share/2302a275-be6f-4397-958d-53cb50f35c68
Unless it has introduced subtle bugs I guess? But it was much faster to identify the key changes necessary and provided the information in a better explained and clearer way. And the code looks better designed at a glance.
Either I’ve already beaten the fallback mechanisms in Fable 5 or they don’t use Opus 4.8 / Sonnet 4.6 / Haiku 4.5 as the fallback. I’ve run several different tasks about upgrading our training library through each model and the output from Fable looks nothing like the others.