The pelican has looked very same-y across all frontier models, same color bike, same camera angle, etc. I suspect this challenge is already too embedded in the training data to be a good signal when it succeeds, and maybe even when it fails in pathological ways mirroring existing AI pelicans on the internet.

Was it ever a good test? How do you even objectively assess what a good pelican on a bike is anyway?

SVG generation is a good test because it's extremely easy to subjectively assess with visual reasoning where humans are strong. However, pelican on a bike specifically may be overused at this point.

The "big beak!" comment in the svg source makes me think it's definitely a gamed "benchmark" at this point.

Do you think the models are ready for the next level? I believe that would be: Pelican feeding Spaghetti to Will Smith.

Variations of this comment have been posted for over a year. The pelican has now morphed into part of HN culture rather than a legitimate benchmark, but it's still valuable as a meme.

it is more an example of gaming (the HN system) than meme.

I'd be very surprised if this is in the training data given that most models mess it up to this day. E.g. look at the ones from Opus.

I'm beginning to wonder how much of a useful metric the pelican is because surely the frontier labs must be training their models on pelican-artistry because of how well known your test is now?

Simon has addressed this on virtually every new model release. He also has unpublished alternate prompts. But the larger point is: this is a fun experiment, not a serious and objective benchmark.

But what if they are better at flamingos? Are they optimized for pelicans? How about “draw me a four headed owl”? The meme, I get it, but I’d settle for a working bash script, tbh.

I just run my own benchmark for "draw an SVG with $animal driving $vehicle". I won't post my choice of animal and mode of transport, but there are plenty of uncommon combinations to choose from. So far it's a fun and visually intuitive benchmark that does seem to correlate with model capabilities

I don't know. Just looking at the bike frames (specifically the fact that the AI generated bikes have rather unsteerable front forks), it's clear to me that frontier labs aren't spending much time tuning models to make bikes look coherent, which I assume is an easier task than making a pelican riding a bike look coherent.

I've seen this reply to Simon's benchmark for 2 years running now, and yet you still see improvements and objectively-bad results over time from new releases, even when I'm sure every frontier AI team has/had a person at least partially dedicated to better bicycle-pelican SVG outputs. Alas.

I had intended to caveat that: I'm sure I'm not the first person to ask about this!

> you still see improvements

This is expected if they are training their models on it, right?

> objectively-bad results

Keen to learn when this has been the case, i.e. across version increments in major models.

I've written about this a couple of times, most notably here: https://simonwillison.net/2025/Nov/13/training-for-pelicans-...

I've been enjoying seeing how the quality of individual models differ based on the amount of reasoning effort you give them. If they were baking an a good pelican you wouldn't expect them to differ so much.

(Google Gemini are the only lab that have very clearly paid attention to the quality of SVG animals-riding-vehicles, see their announcement for Gemini 3.1: https://twitter.com/JeffDean/status/2024525132266688757 )

Amazing, thank you Simon! Look forward to reading.

Hence it has become a meta-benchmark of relative progress in SVG image generation of a known target which has leaked into the training data and for which "every frontier AI team has/had a person at least partially dedicated to" at least checking if not optimizing.

I honestly assumed their comment was tongue in cheek humour, because positively no one actually cares how these models generate an SVG pelican riding a bicycle. It's some meme thing that this stuff always appears here.

Yeah this is not a real benchmark, it's just a fun tradition everytime a new model is released

"fun" / boringly predictable meme thread with 30+ replies already

It is telling that people need to create throwaway accounts to criticize simonw's behavior in this website.

It's evolved from a funny, unserious benchmark to a tradition. When a major new model is released, I now always check the HN thread for Simon's Pelican post. I'll be sad when I don't find it.

When it started, comparing the progress between models was mildly interesting but everyone (including Simon) acknowledges it certainly leaked into the training data long ago.

The way I see it the benefit of benchmark isn't to take Simon's results at face value. It's a template for your own benchmarks that are easy to visually evaluate.

It was a completely useless test even before the labs trained for it.

Yes, it's always been published as a joke. You've explained why it was (and still is) funny meta-commentary on AI benchmarks.

This is the reply I look for in all the new model announcements. Its fun to tell people that I judge models based on pelicans.

Now someone post the link about how it’s impossible for humans to draw a bike from memory.

This is all we need, that moment the Pelican put the leg behind the frame, we are all doomed.

I also look for this reply because i like seeing the follow-up reply saying that this is not a benchmark anymore because labs have gotten it in their training data.

that reply never failed to come it's basically a meme at this point

It's interesting that they still get the head tube / handle bar part wrong.

Or the hands not being wings

I find it quite interesting that while the picture looks better the more advanced the model is, but apparently none so far "understands" that the pelicans legs are on both sides of the bike / top bar.

If you scroll to the bottom of the Fable-5 by effort page, Max effort actually gets this correct! (Along with being the only one I've seen so far to make a bicycle frame that matches the shape of what most bikes on Google images look like)

And the only one linked here that includes a bicycle chain!

How much money do you think they spent fine-tuning on pelican SVG generation?

Not as much as Qwen, since apparently 3.6 35B surpassed Opus 4.7 https://x.com/simonw/status/2044830134885306701

Probably none. They probably have much better targets to optimize for than an SVG pelican or even SVGs in general.

The Max version gets more details right. The bike frame looks good, the chain, the wings are appropriately styled instead of “arms”, and the knee is bent, etc. Obviously we’re hitting marginal returns now, but I see differences.

Where is the clear improvement on Fable 5? The tail is misplaced.

Looks like Fable constructed the "max" "looking" pelican of the previous model for the "xhigh" output token count of the previous model.

Can you please compare the code generated by other similar quality pelicans by other models. Code in your first link (Fable 5 Default) looks minimal yet very good.

It's interesting that Gemini 3(.1?) Deep Think is still the best at this task and it's still not really generally available. Maybe Fable could match it at higher effort levels? https://simonwillison.net/2026/Feb/12/gemini-3-deep-think/

Is it possible to use the credits from subscription (https://support.claude.com/en/articles/15036540-use-the-clau...) for fable?

It also does A LOT better, for my hamster test: https://aibenchy.com/showcase/?q=claude#showcase=6efb87c28e3...

I'm pretty sure they're optimizing the models around these sorts of tests.

I could be tripping but I’m sure that is very similar to the Deepseek one from not long ago. Clearly I am too lazy to go and find it for verification.

Anyone care about these pelicans that always come up anymore?

Clearly at this point they are part of the training data.

They even all look sort of ish the same. Daytime, colors,...

Without being mean, I encourage you to go look at some of simonw's writing on this topic, which he has addressed repeatedly (and IMO satisfactorily.)

I know because I too had this initial take; however, upon analysis, it is not sound.

I know he is an AI influencer that promotes his blog any chance he gets.

I agree as well that he writes many interesting things.

The way they talked it up, having both legs on one side of the bike is like walking to the car wash

Personally feel like it could be more ambitious with what it creates.

Yay, max level actually put one of the legs behind the frame!

Why always sunny days?

Pelicans hate biking in the rain (as do I).

Fable 5 xhigh actually looks the best to me.

Do we need a pelican every single time a model is released? Beating a very dead horse.

Fun at first, seems disingenuous now. A site funnel

that's a great looking pelican

need more Alex Moulton style bikes

dude, the max version looks like it's finally there. handle bar holding with wings, the left leg is behind the frame while the right is in front of it (correctly).

well done anthropic.

mediocre pelican. very disappointing

How many barrels of oil are burned per pelican at Fable levels?