A few notes on Anthropic's new Claude Fable 5.1 - with Max thinking level I got the best SVG pelican I've had from any Anthropic model (at a hefty cost of $3.30!), which I then had it animate

(The wheels running backwards appears to be an artifact of the conversion to video - the SVG has them going in the correct direction:

I got this one with Ultracode

the new watermark is obvious

we all know this benchmark needs to move towards a bicycle riding a pelican

I did enjoy this variant

Jeez, was that only april of this year?

Wheel spokes aside, I’d say this is solved.

You’re clearly in the training envs now Simon

The only eval I trust!

I think this is the first time we got a working pedaling animation? I was waiting for this day. Solved, onto the next

5.1 is actually a huge leap. I was shocked about how good it was. Then it decided I should be speaking with Opus 5 and the whole thing sucked. Anthropic should not be our 'daddy' and just decide what model we need. If I am willing to spend tokens, don't give ne the retard model.

$3.30 for the best pelican is the most simonw expense report

From the animated svg, the wheels are spinning just fine, I think this could be a sample rate illusion!

I’m convinced there is a pelicansvg bench internally at these labs 😆

Question. do you have a gallery where you keep all of these? I think it would be interesting to see how it's evolved between models.

Man, at the rate you're going, the bike riding Pelican can become its own animated series!

That pelican is stick figure like and definitely not “good”

They used to look like this!

I love the fish bouncing in the basket! This is impressive work. I think the pelican cyclist now belongs next to the otter on a plane.

that basket has fish in it, someone at anthropic cared way more than they needed to

Pelican bench solved ✅

$3.30 for a pelican is expensive but that's the frontier model tax. interesting that the wheels reversed in video conversion though

that floppy fish in the basket is adorable! AGI is here!

max thinking clearly does something here, curious if flash tier gets anywhere close on the same prompt

$3.30 for the best pelican is a very Simon benchmark and somehow still useful. Max thinking is clearly where the money goes. Curious how often you leave Max on for real work vs only for the weird SVG tests.

I think the wheel is rotating anti-clockwise, should be clockwise? Doesn't look like there is a rendering sampling issue.

$3.30 for an SVG pelican feels like a metaphor for where we are rn. the capability is there but cost-efficiency for practical work is still the bottleneck. been getting good results with local models for a fraction of that. what's your threshold for 'worth the cost'?

I want this one in SVG now 👇🏼 lessgo! 😅

I made this inspired creation, super impressed with how well the model understands the world one shot

Von Neumann would be impressed that you got it to pedal

It’s been a pleasure to watch this eval saturate

That’s a $100 pelican

@Vjeux Is this really needed ? You are wasting your talent on this outdated thing. Isn’t it :)

@JamesTimmins do your SVG thing

Love following the pelican’s progression 🙂

$3.30 per SVG works for interactive dev sessions, but running 400 autonomous security audit loops a day with full reasoning tokens burns $1,300 before lunch. We isolate deep thinking models to root-cause synthesis and keep sub-agent execution loops on lightweight workers.

this one is solved

Damn if the wheels were turning the right direction this would be perfect