alt Hacker NewsClaude Fable 5
System Card [pdf]: https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c3...
Comments
"Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage"
What does it mean? That they have to add "safeguards" not do erase user disc, or, conversely, they are telling the audience that this model COULD be made so powerful to do some crazy stuff that can hurt governments, etc.? Are they showing off or threatening that if government X would not purchase the license the adversaries might do and what's then!
Just another "a" and we have it. https://faable.com/
> During early testing, Stripe reported that Fable 5 compressed months of engineering into days. In a 50-million-line Ruby codebase, the model performed a codebase-wide migration in a day that would otherwise have taken a whole team over two months by hand.
How in blazes do you end up with a 50M line Ruby codebase? WTF?
jjj
Another Anthropic release, another doomsday for developers.
This time looks like we will only be able to find work making bioweapons, or distilling models.
mankind has reached its final destination
Paging senko, let's see Fable's oneshotted RTS!
The comment under this kind of post is unreadable now. Yeah, probably with 100B you can hire anybody to call something "a beast".
The subscription bit makes no sense has capacity appeared for these 2ish weeks out of thin air that'll vanish? why is it available now but wont be in 2ish weeks?
am i missing something?
why would I pay 200 out of pocket and then some for the best model, it seems very silly.
Anthropic sucks. but this paragraph should be in the "annals of AI-aided self-inflicted learned helplessnes":
> If Claude gives me poor or incorrect advice while I’m working on an AI component, I have no way of knowing whether the model was confused, whether my problem is unsolvable, or if some invisible policy restriction quietly kicked in.
Have you considered actually learning the theory, spending some time actually reading the papers and latest books, paying careful attention even to the eventual math here and there?
Curiously nothing on DeepSWE and ARC-AGI-3 yet. For ARC at least there's a statement that Anthropic won't guarantee them that their secret private test data won't be collected by them and used for training.
Holy shit. I gave it the first actual task I’m facing, it makes me so angry. It just does 7 things more than I asked it fore and it does it so bad. It took 5 minutes and 5 seconds just running time, plus giving me frustration and make me lose my context. Hand-coded I would’ve been done in 3. And it would be code I understand can look at in one year and work on again.
It’s really tough to have sanity fight against hype bros in your head. Probably I should just not visit the internet anymore
To me it’s all just people getting scammed better. With every model it looks better, but it’s at least equally worse to work with, which is the reality it needs to be. It’s less scalable more, code, tougher to understand. Your digging your own grave better kind of.
What pisses me off is that everything people are doing is so walled garden / closed source. Sharing knowledge between companies would be so fucking useful to humanity.
Can we please stop with the extreme "safeguards"? I don't want to waste processing power on a model deciding whether is can answer my question, or ensuring that it's answer is politically correct.
so should I use it with workflows?
"tell me about biology" -> "Switched to Opus 4.8"
this is good
Not included in Max plan. In CC:
> Included in your plan limits until Jun 22, then switch to usage credits to continue.
Seems this will only be available to the 100/month+ folks
I wanted to test the capabilities of the low one, hoping it would be good enough.
I have a quizzes application, and my quizzes only supported flashcards (implemented via table inheritance to provide flexibility for other types of quizzes).
The entire repo is handcrafted, never used any ai on it (it was more of an excuse to test elixir and write code by hand).
Since fable 5 got released the moment I was done with some work, I decided to throw at implementing multi choice questions.
After all it had only to copy the flashcard approach across ui/routing/db, and only had to create a table for the multi choice questions and one for the answers enforcing that all quizzes had one correct question. I told him it had access to sqlite3, chrome mcp for testing and mix commands.
I did a test for low, mid, high. Repeated it twice each.
low-1, and low-2 failed both. In low-1 the UI for adding another choice answers was broken. In low-2 it failed with some unique constraint. It took it 4m36 and 3m59.
Both mid-1 and mid-2 succeeded without issues also implementing the correct ui. They both wanted to use dash at all times. They both wrote tests for the "controller" (or context how they call it in Elixir). They both tried to use the repl to test the behaviour of the schemas.
10m and 12m39.
High didn't demonstrate much gains over mid for this kind of task, it was simply too easy. Times were comparable to mid, but interestingly it used much less bach, and read way more files. Token usage was almost twice the other ones.
But here's the interesting part: I went back to low and added to the prompt two bullet points, to write tests for the controllers and to test the entire flow with chrome mcp.
It produced the same output as mid or high just by adding two instructions to the prompt.
Fable 5 beats GPT 5.5 in my proofreading benchmark. And it does so at approximately the same total cost; it used significantly fewer turns than 5.5
Used it for simple task and I got this message.
Fable 5's safety measures flagged this message. They may flag safe, normal content as well
Being unable to use this with zero data retention makes this feel like a non-starter for most enterprise customers.
Fable? Fabelstories? (Fablestories, but the german word seems more poignant ... Fabelgeschichten ... Fabeln)
Clause 5 ran out of quota with TWO PROMPTS.
Lets let that sink in.
Before long, we'll be having Claude Cylon-class models.
I have been using FABLE 5 with Claude Code since the morning. The speed is very close to what Opus 4.5 was, and the quota use is nearly identical to what it was before the "doubling". Whatever I was experiencing 4-5 months ago is back. Maybe the model is better, but we will see. I cannot tell the difference yet.
my pet conspiracy theory is this is the Opus 4.5 from a few months ago which was extremely good but dumbed down after a week because it was just too good, they didn't want to release it to public. They pulled it down and deployed another "Opus", after that it was just a downhill. Opus 4.8 is unusable for me in React Native, TS, Rails development work.
Opus 4.8 gets stuck in weird loops where Codex one shots the bugs.
I actually rather like the way they have approached these safeguards. Rather than only teaching the model to refuse a request, or completely rejecting the request, the system gracefully degrades to slightly less powerful or slightly less precise operation. So you still roughly have Opus 4.8 even when safeguards trigger, but with an upgrade when they don't. As much as I hate the way they hype Mythos 5, I think the release of Fable 5 is rather nice. What's not nice though is that they plan to remove it from subscriptions soon, but getting to try it is cool, I suppose.
Pelican guy ! Where are you ? :)
AMilliPay.com
Anthropic, can you please stop the FUD?
Release your best model, let the world adapt and evolve, and let's move to the next thing.
It won't even run a basic /security-review command without reverting to Opus 4.8. Utterly useless.
At this point Anthropic is a pure marketing and PR company. Super catchy names like Opus, Mythos and Fable trying to get you to think that these software products are actually super-human life changing experiences. Boris Cherny coming to HN “Hi! it’s Boris from the Claude Code team” to get real tech people’s goodwill.
From Opus 4.6 there are no noticeable improvements for me in code generation. It works very well, till 90% completion, if you guide it correctly. And you need a little luck. For serious production code I need to understand what I’m doing so it helps a bit, sometimes.
is it just me, or this model is simply not available in cc?
the opus 4.8 I assumed wasnt available to enterprise seats, but it explicitly says cc that fable is available in cc. I can't find it, and im on latest version.
is it smart enough to know not to walk to the car wash?
I'm happy not using llms because I like learning things and working hard. I love writing code, it's genuinely my favorite thing thing to do.
Using llms is the equivalent of driving to the store that's 3 blocks away, just like how that's bad for your body (if done all the time), using llms is as bad for your brain.
Before LLMs, we started relying on certain technologies like Maps apps to navigate, now people can't even get around their own town without having access to various cloud services. The implications of not being able to work, think plan without access to an llm are really bad. Its going to destroy your brain and make you an incredibly average person at best.
LLM people are going to lose the ability to read and think for yourself and then your competency is going to be 1:1 correlated to the quality and quantity of tokens you can afford, or a billionaire is willing to allow you access too. Your work will be the mean (at best), because it will the same quality of output everyone else is capable of.
This is seriously the biggest trap by tech. Your bargaining power for your labor is going to get drastically reduced because you won't be able to differentiate your value from anyone else that has access to an LLM. What happens when everyone has the same skill level for certain work? Idk, ask McDonald's employees how replaceable they are. Use them wisely (or not/hardly at all) don't drive to the store 3 blocks away for every little thing you need.
system card = marketing material with heavily gamed benchmarks.
New chapter
This i
Maybe at this point, Fable the game will be played generated by AI as we go.
btw in claude code