Top
Best
New

Posted by Jowsey 23 hours ago

Kimi K3.1 Pro and K3.1 Pro Highspeed(www.kimi.ai)
https://www.kimi.com/code/docs/en/kimi-code/models.html

https://artificialanalysis.ai/models/kimi-k3-1-pro

https://www.modelcompare.ai/blog/china-open-source-ai

1005 points | 603 commentspage 11
dlphii 18 hours ago|
[dead]
cromka 21 hours ago||
Can someone please tell me what benchmarks to look for that measure world knowledge? For sspitballing, critical analysis, business projects, etc. Is the model size the only way to measure it?
yshamoun 21 hours ago|
AA-Omniscience
jampekka 20 hours ago||
AA-Omniscience is done with search off though which is not necessarily a very common scenario nowadays.
mehdi_zare 18 hours ago||
[flagged]
hilde574 20 hours ago||
[flagged]
randbyte 21 hours ago||
[dead]
simonw 22 hours ago||
[flagged]
jorgen123 22 hours ago||
I think I have yet to see a pelican where the left leg and pedal are drawn on the left side of the bike.
simonw 21 hours ago|||
There have been a few that get that right - most notable Qwen 3.8 27B running on my laptop: https://simonwillison.net/2026/Aug/16/qwen-38-27b/#the-defau...

Also the recent Geminis - https://simonwillison.net/2026/Oct/13/llm-gemini/ - and GPT-6 Astra on Max: https://static.simonwillison.net/static/2026/gpt-pelicans-gr... - and Claude Fable 5.1 https://static.simonwillison.net/static/2026/claude-pelicans...

CamperBob2 21 hours ago|||
GLM 5.3 not only got that part right, but animated it: https://crimson-jeri-74.tiiny.site/

To be fair I have a feeling that if you added 'animated' to the SVG request, the output quality from many other models would improve, just because of the extra chain-of-thought work needed to fulfill the more complex request. IMHO the animation requirement should be added to the canonical pelican prompt.

minimaxir 21 hours ago|||
Simon's comment apparently got flagkilled; I vouched for it.

If enough people are flagging it, we may be at a point where the pelican benchmark is truly overdone.

gnabgib 18 hours ago|||
Please do not decrease the useful signal with this noise
mi_lk 19 hours ago|||
Is that even a debate at this point? Doing my job as a netizen to maintain signal/noise ratio
tetraodonpuffer 22 hours ago|||
definitely not frontier though, as even on max both legs are in front of the frame; although I am wondering how much getting that right correlates with model intelligence.
simonw 22 hours ago||
The biggest (of many) failing of the pelican test is that it's only one prompt - what matters most these days is how well a model can drive a lengthy agentic tool calling session.
MrBuddyCasino 21 hours ago|||
I like the low effort one most.
ciefa 22 hours ago|||
Can I just say that I love how you are also just always here the moment something drops lmao. Hope you're all well! The pelicans indeed are solid.
dude250711 22 hours ago||
What is the primary distillation source?
henry700 22 hours ago|
you
Super3000 23 hours ago||
I'll wait for the pelican benchmark.
fenderog 16 hours ago|
here's a fun benchmark https://fenderog.com/art-crit/page.html?p=ceate-an-svg-of-an...
doctoboggan 22 hours ago|
I always assume they are benchmaxxed until proven otherwise.