Voxelbench

A creative benchmark for evaluating language models on making voxel builds.

[Voxelbench @voxelbench]

Voxelbench - our new benchmark for evaluating AI models on voxel builds!

[Voxelbench @voxelbench]

Gemini 3.6 Flash is now on VoxelBench

[Voxelbench @voxelbench]

correction:

For the 4th example output at the end, the alt text for the Build request should say:

"Kimi K3 (Max): Create a scene from the future"

[Voxelbench @voxelbench]

https://t.co/wWHdShpLsw

[Voxelbench @voxelbench]

Kimi K3 ranks 3rd on VoxelBench

just 100+ Elo points behind Fable!

and a huge uplift from K2.6 (28th)

[Voxelbench @voxelbench]

Kimi K3 lands on VoxelBench

The outputs look great so far!

Full benchmark run still ongoing ⏳️

[Voxelbench @voxelbench]

Claude Fable 5 max

"Om Nom playing a level in Cut the Rope"

[Voxelbench @voxelbench]

gpt-5.6 sol on max effort

It can "write" really nice in voxels

[Voxelbench @voxelbench]

@nicdunz We will try to post more gens frequently

[Voxelbench @voxelbench]

GPT-5.6 Sol overtakes Fable 5 on VoxelBench

the highest scoring model yet with over 1k votes!

[Voxelbench @voxelbench]

A first impression of GPT-5.6-Sol (Max)

I think the picture speaks for itself.

[Voxelbench @voxelbench]

Grok 4.5 hit the leaderboard!

Impressive step forward from @SpaceXAI!

[Voxelbench @voxelbench]

SpaceXAI has released Grok-4.5 and it is a huge improvement over previous Grok versions.

It matches Opus-4.8 in benchmarks.

Visit us at voxelbench to see more outputs from Grok 4.5 or vote to shape the leaderboard!

For Elon, we ran 2 Mars related prompts.🚀

[Voxelbench @voxelbench]

Claude Fable 5 (Max)

> A Rabbit in a dress shirt and blazer pointing at the minute hand of a pocket watch.

[Voxelbench @voxelbench]

Claude Fable 5 on Max effort

> Atlas and P-body making a dramatic pose

[Voxelbench @voxelbench]

Fable is back! And we totally understand the hype!

It tops our leaderboard by a good margin.

[Voxelbench @voxelbench]

Fable 5 and Sonnet 5 are now on VoxelBench!

Check out their builds and vote!

[Voxelbench @voxelbench]

We love GLM-5.2 too!

[Mirochill @mirochill]

🤯 Claude Fable 5 vient d’être ajouté à @voxelbench !

J’ai regardé quelques générations… et franchement, elles sont toutes impeccables.

Il pourrait bien dépasser GPT-5.5, actuellement en tête du classement 👀

[Voxelbench @voxelbench]

@_Suresh2 In the near future we will show the total costs