if you tried hyperframes, you would know the agentic video stack is being built on code-gen but SWE-bench code v.s. code-to-video that feels alive are two different things we built the Code2Video Bench, in collab w/ Google DeepMind & Kaggle frontier labs can finally get good at agentic video tasks

Read the full report here:

the dataset and most up-to-date ELO leaderboard on @kaggle:

@kaggle this space moves ridiculously fast

@kaggle especially for something this subjective

@kaggle This could raise the bar across the board

@kaggle straightforward, public, human-grounded

@kaggle let's gooooooo

yep

@kaggle Nice

@kaggle public leaderboard season begins

168 briefs is commitment

@kaggle This feels more like testing creative judgment than testing raw capability

@kaggle we need more benchmarks like this

@kaggle Which video would you actually choose. simple

@kaggle Finally a benchmark normal people can understand lol

@kaggle the best benchmarks make progress obvious

@kaggle This makes so much sense.

@kaggle now that would be cool

@kaggle this gives creative quality something concrete to compete around

@kaggle This is exactly the kind of benchmark the space needed.

@kaggle A benchmark for whether ai video actually feels good. finally

@kaggle honestly that's exactly what was missing.

@kaggle 82% agreement with humans is impressive work

@kaggle Very very cool

@kaggle no need to overcomplicate it

@kaggle not just another leaderboard trophy

@kaggle Ai video getting a real report card now

@kaggle Very nice

@kaggle those edge cases are probably fascinating.

@kaggle finally

@kaggle Interested to see what builders do with it

@kaggle very interested in the cases where the judge and people disagree.

@kaggle "Code isn't just about syntax anymore; it’s becoming the ultimate rendering engine. Code2Video Bench might just be the playground where the next generation of AI directors is born."

@kaggle finally we can see how good these models really are at code-to-video 🔥

@ElizabethAI_1 @kaggle love this

@kaggle Yet here we are 🤯