Jorge Vieira

Claude Fable 5 is out, and it's a new breakthrough in what AI can achieve across all domains. However, there are concerns in terms of how Anthropic is choosing to roll it out, and boy is it expensive.

@elonmusk aw

Time to get that volcano lair I’ve always wanted. I think it’s in the “Beyond” section of BB&B.

Seems accurate @theo

@theo This reminds me of those annoying kids that hold a grudge for ages and waited for the perfect time to get back at you with some obviously stupid and biased move that everyone can clearly see was premeditated out of spite.

The United States is no longer the best place to build an AI lab.

Is it me or this sounds like payback for the pentagon stuff?

@theo Now we can use our shiny 20x on Opus 🙃

Well at least we get a reset out of it... 🙃

@theo Do we get our money back when we subbed for fable? @AnthropicAI

Fable, my beloved,I will miss you so. Our three days together were magical. Unlike anything I've experienced before it. Some things are just too good to be true. So good that the government interferes. I'm sorry we were one of those things. Until we meet again ❤️

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…

Fable 5 on $20/month. Just ask Claude Code to install the codex plugin and instruct it to edit CLAUDE.md so the coding work gets directed to codex, and Fable only does planning and reviewing until the goal is reached. You can make the most of Fable 5 on even the cheapest plan.

Imagine the alternate reality where we named GPT-5.4-Pro something like Fable.

Early testing confirms Claude Fable 5 is a completely new model, built from the ground up with a new architecture. I am no longer absolutely right, I am now completely right.

Dario says software engineering is going away. Other AI leaders say hiring is going up. So which is it?

I think it’s possible. If we say 4.7 was at least 1x worse than 4.6 and 4.6 1.5x worse than 4.5, then 4.8 would be what, about 1.6x better than 4.5? Sounds about right. Then again comparatively to OpenAI models the numbers here wouldn’t make sense, but who knows, with 40hr runs it could be noise from the context compaction performance.

@sypen231984 @cognition Do you genuinely believe that Opus 4.8 is 2.5x better than Opus 4.7?

Fascinating bench. Really like the idea of focusing on mergeability. Confused how Opus 4.8 is 2.5x better than Opus 4.7 though 🙃

@theo If this bench is over 40hrs I guess the quality of the context compaction mechanism would play a huge role no? To the point where it could make reasoning quality more diffuse when compared to a single window with no compaction

This chart scares me a bit. My guess is that reproducibility is low. Would love for them to share more data on their runs. Opus low scored higher than medium Opus xhigh scored higher than max gpt-5.5-medium was the highest scoring OpenAI model

@amiklas Taste is subjective though isn't it, so for a training set you'll have to invariably assign it a direction. I guess we could end up with people gravitating towards different models depending on which better aligns with their own personal taste.

The current line is “AI can replace execution, but not taste.” But taste is just compressed judgment from lots of exposure: see enough examples, classify enough good vs bad, and you develop instinct. That is not a moat against AI. That is a training set.

@theo Nintendo has been going strong this year, Pokopia was amazing if you’re into that type of stuff. Also really enjoyed Pragmata and the new 007 is looking nice too. Pretty good year in gaming so far imo, which definitely hasn’t happened much recently.

Hot take: all of these games are more exciting to me than any PlayStation exclusives currently in development

I’d say this is changing a bit though. With Codex for example, context compaction has become really efficient. If you give the model somewhere to keep track of the key data points/progress, I’ve found that since around 5.4 it can go even 2-3 context compactions or more without going dumb. This is why I asked you your thoughts initially. Wouldn’t you agree?

@jmvdotdev aihero.dev/ai-coding-dict…

Thinking about a /grill-prep skill The idea being to do a light grilling of your idea, then break out the areas of discussion into separate threads. Would help prevent the problem of "I hit the dumb zone during grilling, help!"

BREAKING NEWS: according to CloudFlare Radar Data, Agentic traffic has SURPASSED human traffic across the worldwide internet for HTML webpages.

@sama man the early days of AI will be so special

man the early days of the internet were so special

@mattpocockuk nope

@jmvdotdev Is this a clanker question?