Thomas Walker

[Thomas Walker @ThomasMWWalker]

This should be a new type of recipe book

[juan @juanbuis]

ok 31k of you bookmarked this so i built it

paste any recipe, or just a link, and it becomes one of these

https://t.co/p5T6zoU0Ve

[juan @juanbuis]

i can’t believe someone solved recipe design decades ago and everyone just decided to ignore it

[Thomas Walker @ThomasMWWalker]

How I feel using a keyboard after talking at Claude and Codex for an hour

[Thomas Walker @ThomasMWWalker]

@AndyMarlowg Yep, and difference between Opus 5 (left) vs Opus 4.8 (right) is incredible

[Thomas Walker @ThomasMWWalker]

@Max_Brind27 @AnthropicAI Yeah and Opus 5 can be pushed much further than this prompt did for sure

[Thomas Walker @ThomasMWWalker]

@AnthropicAI’s Claude Opus 5 is the best model for 3D rendering!

I gave 9 model/reasoning configurations the same one-shot prompt: build an interactive 3D robotic-hand simulation.

Opus 5 High produced by far the best result, in my opinion. The difference versus Opus 4.8 (and every GPT configuration I tested) was night and day.

Grok Build 4.5 is impressive, especially given its speed and price.

Video 1: all nine side by side
Video 2: Opus 5 High
Video 3: Opus 4.8 High
More clips and link to try it yourself below 👇

[Thomas Walker @ThomasMWWalker]

@mattshumer_ Yep it is incredible. Claude Opus 5 vs Opus 4.8 on the same prompt here:

It is amazing to see how much better 5 is!

[Thomas Walker @ThomasMWWalker]

@chetaslua @threejs Look at how realistic this robotic hand simulation is! This was one shotted from Opus 5 High. It is by far the best model for 3D. GPT Sol High is the second video for comparison

[Thomas Walker @ThomasMWWalker]

@minchoi Agreed. It really is a step change relative to Opus 4.8 (honestly all other models) for 3D.

This robotic hand simulator was done with one prompt: video 1 is Opus 5 High, video 2 is Opus 4.8 High

[Thomas Walker @ThomasMWWalker]

Play around with the sim here: https://t.co/VYDxUnyV4T

[Thomas Walker @ThomasMWWalker]

@AnthropicAI Impressive result from Fable 5, although not quite as good as Opus 5 as the palm coupling wasn't as realistic. It was also much more expensive to generate.

[Thomas Walker @ThomasMWWalker]

Sonnet 5 High is better than GPT 5.6 Terra High. Terra struggled to get the thumb to oppose the hand in the correct way, which also was the case for GPT 5.6 Luna and Opus 4.8.

[Thomas Walker @ThomasMWWalker]

@AnthropicAI One detail that separates Opus 5 from GPT Sol High is palm cupping. As the ring and little fingers bend, the palm should naturally curve with them. Opus 5 captures this coupled movement far more convincingly, making the hand feel much more physically realistic.

[Tobin South @TobinSouth]

bloody ripper

[Theo - t3.gg @theo]

I got so used to the limits with Fable that Opus feels practically unlimited

[Australian Institute for Machine Learning @TheAIML]

This could be the fastest $250 you'll ever make! Submit (unlimited) names here https://t.co/Ky1sP9J6tX. Winner announced 31/07/20. #Competition #AI #MachineLearning @ecms_uofa @UniofAdelaide

[Thomas Walker @ThomasMWWalker]

@presidentdalby I hope to see both rings today @AusSpaceLuke!

[Gabriel @tesla_addicted]

Watching it on YouTube doesn’t feel the same as LIVING it!!!!! @elonmusk @tesla #v10 #teslav10