did fable 5 actually get nerfed after the relaunch
saw people saying the bridgebench scores tanked after it came back online. is this real or just people complaining. trying to figure out if i should keep paying for max
yeah it definitely feels worse after july 1. i use it for generating email drafts and product descriptions and it's way more conservative now - keeps refusing to write sales copy because it thinks im trying to do something sketchy
ok so wait is fable 5 even usable now or should i just forget it exists and use sonnet 5
This tracks with what we're seeing too. We use fable 5 to generate product descriptions and email copy for our e-commerce clients, and after the july 1 relaunch it refuses to write anything that even mentions pricing or competitive comparisons. The safety classifier flags it as potential deceptive marketing. Before the takedown we could generate 40-50 product descriptions in a batch, now we're getting maybe 12 before it starts refusing. Had to switch back to sonnet 5 for most client work.
yeah we're seeing the same thing. fable 5 refuses to generate anything that looks like sales copy now. completely unusable for marketing content
we hit the exact same refusal pattern in our testing environment. fable 5 post-july 1 flags anything with "optimize conversion" or "increase clickthrough" as manipulation. the safety classifier is way too aggressive on marketing/sales language. scored it on 50 email templates that worked fine pre-takedown and 38 got rejected outright
yeah it definitely got nerfed.... the relaunch on july 1 has a safety classifier that flags a ton of stuff. people are posting before/after evals and the coding scores collapsed from 86 to 26 on debugging tasks
honestly not sure it's a nerf vs just different safety tuning that hits certain patterns harder.... would love to see the actual eval prompts people are using for the before/after comparison
do you have before/after eval numbers or just vibes? everyone says it got nerfed but i want to see actual metrics