Astra vs. Fable 5.1: tag, youโre it! ๐๐ฝ
Tag Game, all in real time, in the NYC simulation Astra itself created.
The catch? The world doesnโt pause while they think. Cue some awkward collisions with pedestrians and cars ๐
Frontier agents are stepping into embodied worlds, and a simple game of tag puts spatial awareness, planning, opponent prediction, and timely reactions to the test. Skills theyโll need to thrive beyond the chat or coding window.
Watch Fable 5.1 beat Astra 2-1 ๐๐
Coding agents like GPTโ6 Astra and DS V4.1 Flash are entering the 3D world. We need new benchmarks for their spatial reasoning. ๐๏ธ
Introducing _BuildingBench_: can coding agents turn real-world images into coherent 3D buildings, and at what cost?
BuildingBench is a step toward coding agents creating reusable worlds that people and other agents can inspect, edit, and build on for design, games, and physical simulation.
And the qualityโcost frontier is moving fast: ๐
๐ฅ GPTโ6 Astra achieves the highest score in our comparison at ~80% lower cost than Fable 5.1.
๐ฅ DeepSeek V4.1 Flash delivers strong quality at a median model cost of just $2.31 per building.
Watch the frontier shift. More details, interactive leaderboard, and GitHub in the thread ๐
I increasingly think future software will combine code for control flow with small neural programs for "fuzzy" judgments.
That's what I've been exploring with ProgramAsWeights, which answers the question of where those neural programs come from: they are "compiled" from English descriptions.
For example, I combined 30 neural programs with a decision tree to build a course website helper that looks like a chatbot but runs locally. Each small program handles a question like "Which specialist answerer should this be routed to", and code controls the overall flow.
You can try building with it here:
https://t.co/Bwgcn362bc
GTA 6, the GPTโ6 edition! ๐๐ฝ
Added pedestrians & traffic to our earlier NYC city built by Astra.
Plusโ๏ธthe player is controlled by Astra too! It explores NYC on foot and by car, a bit rude to the pedestrians, though ๐ ๐
Astra built the world. Now itโs playing in it, for about $0.40/min.
What should its next mission be? ๐
We built an interactive 3D heart with GPTโ6 Astra. ๐ซ
In 8 hours, Astra read 8 research papers, reviewed anatomical atlas and microscopy data, and interacted a bit with a human expert, to build 3 visual layers: anatomy, blood vessels, and nerves.
8,100 modeled vascular and neural branches, plus detailed valve leaflets, papillary muscles, and their supporting chordae.
From cities to hearts. Hyper-realistic simulation is showing promise in both physical and biological worlds.๐ซ๐
Interactive demo + papers and data sources in the thread ๐