Huge fan of Claude for a long time but switched now to Codex after all of this "ok one more week, one more week" stuff. It feels like Sol is quite capable, and the nice thing is the refusal rates seem more reasonable, _especially_ if you're doing anything thats reasonable but that they consider "sensitive". The dealbreaker for me is their initial nerfing / silent sabotage but now down-grading for anything even resembling AI work. That's a pretty low blow IMO.
Limiting the amount of Fable usage makes us reserve Fable for our hardest problems, which is probably a very important signal when it comes to how valuable the training data is.
In other words, I think they've found a way to manipulate their customers into labeling sessions as "high quality data" (Fable sessions) and "low quality data" (non-Fable sessions).
Fable is insanely annoying. It treats me like an incompetent, suggesting when something is in my expertise that I'm actually wrong. And then it will simply refuse to engage or do what's asked. Anything related to present day American Politics is strictly off limits. How is this truth seeking, or rational? Model misalignment increases when post-training is done on non verifiably domains. Scary times indeed.
Co-worker and I were speculating that a) They'd extend the preview period when OpenAI announced Sol (which they did, maybe not because of Sol though) and b) That once the preview period it was going to get included by default on plans in a couple weeks so they could gauge usage of it being included versus having to pay for it all the time. Noticed that Fable was included when I launched claude code this morning, so I'm guessing they have all the data they needed to make a decision.
Due to all their fumbling I decided to not even try Fable. Why get used to a model with limited shelve life when I can just use the older models that do well for me? Well then 5.6 came out, now Kimi came out and I’m just letting my Anthropic subscription expire. Anthropic will need to do a whole lot better because this just ain’t it.
I recommend people do mini "evals" from time to time, ask the same question to different models, and eventually you'll arrive at the conclusion that there is nothing like Fable.
I was confused about this a few hours ago. Why did I still have Fable 5 on my Max plan today? Am I getting billed per token, eek? (No.)
This is the message in in the claude [code] cli:
Fable 5 is now a standard part of your Max plan
You can use up to 50% of your weekly usage limit on Fable 5. If you hit your limit, you can continue on Fable 5 with usage credits. Fable 5 draws down usage faster than Opus 4.8. Run /model and select Fable to use it. Learn more: https://support.claude.com/en/articles/15424964-claude-fable-5-promotional-access
I've noticed Fable's safeguards being a lot more sensitive today. Like, waayyy more sensitive than they were just yesterday. It feels like they did something...
No? I have no problems with the 20$ plan and I'm using Opus 4.8 on high. At the end of the week I have a few percent left so it fits right in for me. I don't use it for work, only for personal stuff but I've quiet a few private projects.
"Fable is now included on Max plans (up to 50% of weekly limit) if you actually have any usage left after when you thought it'd be taken off you yesterday"
haha, totally the same situation. also, they did an unscheduled limit reset on thursday (less than 24h before the usually planned reset), so had to urgently reshuffle my priorities to have one week of fable squeezed into ~20 hours (and mind the session limit too, as it becomes the bottleneck in that situation)
Huge fan of Claude for a long time but switched now to Codex after all of this "ok one more week, one more week" stuff. It feels like Sol is quite capable, and the nice thing is the refusal rates seem more reasonable, _especially_ if you're doing anything thats reasonable but that they consider "sensitive". The dealbreaker for me is their initial nerfing / silent sabotage but now down-grading for anything even resembling AI work. That's a pretty low blow IMO.
Limiting the amount of Fable usage makes us reserve Fable for our hardest problems, which is probably a very important signal when it comes to how valuable the training data is.
In other words, I think they've found a way to manipulate their customers into labeling sessions as "high quality data" (Fable sessions) and "low quality data" (non-Fable sessions).
Fable is insanely annoying. It treats me like an incompetent, suggesting when something is in my expertise that I'm actually wrong. And then it will simply refuse to engage or do what's asked. Anything related to present day American Politics is strictly off limits. How is this truth seeking, or rational? Model misalignment increases when post-training is done on non verifiably domains. Scary times indeed.
That's what you get when American companies bow to pressure from the federal government in order to release a product.
If you personally offend the President or a member of his cabinet in any way, you don't get to go to market.
Meanwhile China's censored open models are looking more and more attractive in direct comparison to censored American models.
Co-worker and I were speculating that a) They'd extend the preview period when OpenAI announced Sol (which they did, maybe not because of Sol though) and b) That once the preview period it was going to get included by default on plans in a couple weeks so they could gauge usage of it being included versus having to pay for it all the time. Noticed that Fable was included when I launched claude code this morning, so I'm guessing they have all the data they needed to make a decision.
Due to all their fumbling I decided to not even try Fable. Why get used to a model with limited shelve life when I can just use the older models that do well for me? Well then 5.6 came out, now Kimi came out and I’m just letting my Anthropic subscription expire. Anthropic will need to do a whole lot better because this just ain’t it.
If they want to sell me 50% something I don't want then they get 0% sale and my money goes to their competitors. Simple as that, Anthropic.
Just don't use fable if you don't want it?
The next logical thought would be, maybe I want the thing I can't have (i.e. 100% of subscription used for Fable).
Oh, I misinterpreted your initial comment.
Not surprised, considering how bad it is. Even the simplest tasks trigger a safety. I couldn't imagine paying for it extra.
I recommend people do mini "evals" from time to time, ask the same question to different models, and eventually you'll arrive at the conclusion that there is nothing like Fable.
The effort frontier model providers are going through to generate a permanent client base would make Philip Morris blush.
Make no mistake: the only reason you are getting the Fable 5 on Max plans is the Chinese competition.
Without it, you would be paying $2,000/month (Sam Altman, circa 2023/24 when he was saying GPT-4 had IQ of 130).
Hello, Codex
I was confused about this a few hours ago. Why did I still have Fable 5 on my Max plan today? Am I getting billed per token, eek? (No.)
This is the message in in the claude [code] cli:
I've noticed Fable's safeguards being a lot more sensitive today. Like, waayyy more sensitive than they were just yesterday. It feels like they did something...
Earlier: https://news.ycombinator.com/item?id=48975142
Previously: https://news.ycombinator.com/item?id=48954522
...but not for Pro plans anymore, they give you one-time 100$ extra usage credits instead (well it said 100$ but I only got 85).
Max would be a 7x cost increase for me, I think it'd have to do a bit more than it does to justify that.
I'm so sorry you can't have the extremely intelligent and expensive model for $20
Why so snarky? They included it in my plan "for free", I tried it, felt that it isn't worth the price at this point.
$20 Pro plans are still BS starter plans anyway, right? I blew through that in an hour when I first tested out Claude Code.
No? I have no problems with the 20$ plan and I'm using Opus 4.8 on high. At the end of the week I have a few percent left so it fits right in for me. I don't use it for work, only for personal stuff but I've quiet a few private projects.
Oh ffs.
Title should be:
"Fable is now included on Max plans (up to 50% of weekly limit) if you actually have any usage left after when you thought it'd be taken off you yesterday"
haha, totally the same situation. also, they did an unscheduled limit reset on thursday (less than 24h before the usually planned reset), so had to urgently reshuffle my priorities to have one week of fable squeezed into ~20 hours (and mind the session limit too, as it becomes the bottleneck in that situation)
Disappointed Kimi has closed subscriptions due to capacity constraints.
With Kimi and Codex there is no need for silly Fable.