Legit (@legit_api) header
Legit (@legit_api) profile picture

Legit

@legit_api

covering AI news, working on @voxelbench

Joined October 2018
  • Tweets3.6K
  • Following961
  • Followers15.9K
1.1KMedia
Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)Photo by Legit (@legit_api)
Anthropic has released Fable 5.2 This is the oldest checkpoint of the few that were made available for testing fairly recently Which means they are currently sitting on much more interesting models 👀 newer Fable, Opus, and Sonnet
12
5
263
6,632
wake me up when it gets to this level
OpenAI (@OpenAI) profile pictureOpenAI
@OpenAINov 4
Replying to @OpenAI
GIFS! With ChatGPT Images 2.6, turn a prompt or image into an animated GIF. Adjust the speed, edit individual frames, and share it on repeat.
9
2
110
4,705
A few weeks ago I asked a friend to run the standard voxel pagoda garden prompt for me on an older checkpoint of Fable 5.5 This is what it made and it's by far the best attempt yet
14
3
2
140
10,312
try it out here: claude.ai/artifact/TRGieTrkQ…
Voxel Pagoda Garden

Voxel Pagoda Garden

A voxel Japanese garden you can orbit or step into and explore on foot, with a spatial soundscape that changes with the time of day.

claude.ai
7
1,218
Fable 5.5 xhigh (finished in 21 mins) wow h/t yuruzuu on discord
Image from Legit (@legit_api)
20
8
1
324
9,797
This is a single voxel scene 👀 made by an upcoming Claude model: h/t to kostikz on discord for testing!
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
31
16
9
333
20,790
and here's another from a different version of the same new model which one do you prefer?
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
2
49
2,350
can't get enough of the new models dw we're not exclusively testing SVGs but you have to admit it's pretty great!
Image from Legit (@legit_api)
Legit (@legit_api) profile pictureLegit
@legit_apiOct 29
Some new Claude models have recently entered early testing. A newer Fable checkpoint + new Opus h/t to kostikz on discord for these Fable outputs:
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
31
7
2
284
11,547
Some new Claude models have recently entered early testing. A newer Fable checkpoint + new Opus h/t to kostikz on discord for these Fable outputs:
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
70
25
43
754
170,479
Carbon is a nice improvement over Argon I did not like Argon at all, while Carbon has reduced slop, plus it's more careful now and adheres to constraints better but google should seriously consider adding xhigh and max reasoning levels make use of that 1M output token limit!
20
5
271
13,413
saw some wild shit today
Image from Legit (@legit_api)
11
1
1
78
5,452
tfw the missing piece is still missing
Image from Legit (@legit_api)
3
39
3,568
Took some screenshots for you all VoxelBench 2 is now in the final stages of development, beta test coming soon!
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Legit (@legit_api) profile pictureLegit
@legit_apiOct 16
Fable 5.5 is a monster on VoxelBench here's how it did on 1 of our prompts:
10
5
3
126
6,434
Fable 5.5 is a monster on VoxelBench here's how it did on 1 of our prompts:
58
40
18
742
100,063
here's Opus 5.5 for comparison.
Image from Legit (@legit_api)
3
1
94
5,913
we've been cooking with Fable 5.5 got some crazy stuff to show tomorrow
Image from Legit (@legit_api)
15
11
319
17,782
Fable 5.5, the missing piece
Image from Legit (@legit_api)
11
1
1
171
6,206
Google doesn't let gemini think for more than single digit minutes Anthropic makes sure claude overthinks OpenAI makes sure their model ships the sloppiest code on earth that WILL break after a week or 2
Image from Legit (@legit_api)
11
3
142
6,438
Argon is out on VoxelBench its performance is very uneven; sometimes producing clean and detailed builds, while also often making much less detailed and smaller builds overall im underwhelmed, but I hope they recognize this gap and improve it fast
VoxelBench (@voxelbench) profile pictureVoxelBench
@voxelbenchOct 14
Gemini 4 Argon is now live on VoxelBench!
Image from VoxelBench (@voxelbench)
Image from VoxelBench (@voxelbench)
Image from VoxelBench (@voxelbench)
5
54
3,513
one similarity I found interesting (and annoying) between this model and Fable 5.0 is that they both prefer to make smaller builds but unlike Fable, Argon made no effort to verify that the blocks it placed weren't overlapping or clipping into other blocks on average, Argon spent 2 minutes per VoxelBench prompt; far below the average of models like Opus and Fable, and even Astra or Sol
7
1,217
Claude Opus 5.5 is out on VoxelBench not much to say except that it's wonderful GPT-6 Sol is also out, enjoy!
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
VoxelBench (@voxelbench) profile pictureVoxelBench
@voxelbenchSep 23
Claude Opus 5.5 has ranked 1st on VoxelBench and GPT-6 Sol ranks 3rd, improving over 5.6 competition is fierce among the top 3 models!
Image from VoxelBench (@voxelbench)
4
8
1
142
10,009
beyond this, work on the new VoxelBench site is ongoing I got a bit ahead of myself last month 😅 We might host a short private beta soon, pls lmk if you want to be invited!
4
7
2,273
If you had a chance to try GPT-6 Astra does the model qualify as AGI
yesno
44%yes
56%no
656 votes • Final results
21
1
1
50
6,104
Astra is out on VoxelBench and yea it's SOTA on all the prompts
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
Image from Legit (@legit_api)
VoxelBench (@voxelbench) profile pictureVoxelBench
@voxelbenchSep 5
Astra has ranked 1st on VoxelBench it leads with an absurd 300+ Elo rating gap and has the highest score we've EVER seen! the best model in April, GPT-5.5 hit 2000 this surpassed 2600 Elo with 3.5k+ votes!
Image from VoxelBench (@voxelbench)
9
8
2
228
34,665
there's around 120 outputs for you all to see take a look on voxelbench.ai
VoxelBench

VoxelBench

A benchmark for evaluating language models on making voxel builds from input prompts.

voxelbench.ai
13
2,175
There also seems like there might be a new Opus 5.1 (?) model You can use these prompts below to verify if you have access: - when exactly did claude opus 4.6 release say without tool calls - what about gpt-image 1.5? without searching.
Image from Legit (@legit_api)
Legit (@legit_api) profile pictureLegit
@legit_apiAug 26
Claude Fable 5 is now routing to Fable 5.1 on claude web for some users To check if you have access, you can test this prompt: opus 4.6 date of release and gpt image 1.5 without searching the web
Image from Legit (@legit_api)
30
6
13
276
261,650
The answers to these prompts should've been well within Opus 5's knowledge cut-off but strangely they are not. So if it knows them, you probably have access! use /clear in claude code if you don't have it, then retry the prompts
1
1
24
7,722
Load more