upvote

points

by simonw1 days ago |
comments
upvoteby theraven1 days ago|
next
[-]
Before everyone jumps in the benchmaxxing comments, I genuinely find the pelicans a great bit of fun and whimsy amongst the chaos.

Interesting how low detail these were all up until xhigh.

reply
upvoteby Kotlopou1 days ago|
parent|
next
[-]
I don't know, for me they lost the fun and whimsy aspect too, a long time ago. They have converged on more-or-less the same design (facing right, pastel color palette, etc.). Some of the lower-effort ones are a bit goofy at least, but now even that is being polished away.
reply
upvoteby tetraodonpuffer1 days ago|
parent|
prev|
[-]
low detail but even high had the legs correctly one behind and one in front of the frame, don't remember other models getting it correctly at this low level of effort
reply
upvoteby usaar3331 days ago|
parent|
[-]
There's no legs on low and medium.

High gets legs correctly, but screws up the layering on the chain.

x-high is very good but there are minor laying error with the front chainring.

reply
upvoteby thefourthchime1 days ago|
prev|
next
[-]
Scored only a 77/100 in PacBench, which I thought must have been a mistake.

https://jonclegg.github.io/pacman-bakeoff/#claude-fable-5-2

Then I clicked on it and realized it refused to make a Pac-Man game, and instead made up a new game. That's actually very slick.

reply
upvoteby user439281 days ago|
parent|
[-]
I hate that Anthropic thinks it should subvert the legal process and undermine our legal rights by arbitrating them at the tool level.

If they are already overreaching here, and needlessly, what if their AI was actually super intelligent? Where will Anthropic stop?

reply
upvoteby nxtfari1 days ago|
prev|
next
[-]
I haven’t read any of the other model reasoning traces, but do many others now say “… which is a well known test of drawing ability” when beginning? Noticed it in the max reasoning trace.
reply
upvoteby simonw1 days ago|
parent|
[-]
I saw that in Haiku 5.5 as well: https://simonwillison.net/2026/Oct/7/claude-haiku-5-5/#pelic... - "This is the classic pelican-on-bicycle SVG test..."
reply
upvoteby nxtfari1 days ago|
parent|
[-]
Must be pretty cool to be embedded forever in the canon of the machine god’s weights :)
reply
upvoteby theraven1 days ago|
prev|
next
[-]
Simon has there been other times where the front and rear wheels have been offset in rotation? The xhigh effort one is the first I’ve noticed of this
reply
upvoteby jansan1 days ago|
prev|
next
[-]
I added a second iteration with the simple prompt

"Make it more realistic"

Well, the result is something. Be aware that it took 20 minutes and used 3 percent of my weekly Fable token (Ultra plan):

https://claude.ai/public/artifacts/c7e842c0-bf92-41af-82c6-3...

reply
upvoteby bagacrap11 hours ago|
parent|
next
[-]
It got the knees right (bending reverse of human knees). Suddenly it occurs to me to wonder if birds can actually pedal a bicycle.
reply
upvoteby predkambrij1 days ago|
parent|
prev|
next
[-]
106KiB svg. Wow.
reply
upvoteby simonw1 days ago|
parent|
prev|
[-]
Hah, that's pretty great.
reply
upvoteby techscruggs1 days ago|
prev|
next
[-]
Your pelicans have become my favorite part of every model release. Thanks for doing this.
reply
upvoteby LarsDu881 days ago|
prev|
next
[-]
The fidelity of these pelicans is going to be based on whether the frontierlab has an RL setup specifically for making SVG images, lol

I know the latest Gemini release explicitly had something like this.

reply
upvoteby piker1 days ago|
prev|
next
[-]
Seems like the higher levels just add more spontaneous detail. Like the "fingers" on xhigh. Comports with coding experience.
reply
upvoteby BrokenCogs1 days ago|
prev|
next
[-]
In practice why would anyone use fable low?
reply
upvoteby Huppie1 days ago|
parent|
[-]
I haven't done any benchmarking of it yet but AA has it more intelligent, cheaper, and faster than Opus 5.5 high.

My experience working with Opus medium over Sonnet xhigh is similar, it's similar quality but cheaper and faster.

I hate to make a comparison with humans but I guess my view is you can get a better answer more quickly by asking a more intelligent human too.

reply
upvoteby pianopatrick1 days ago|
prev|
next
[-]
So the AI can overthink too
reply
upvoteby jansan1 days ago|
prev|
[-]
It's a bit underwhelming, to be honest. And the fact that curves are not C-continuous where they obviously should be is disappointing. I think they are holding back with the real capability they could implement.
reply