@enactraai
Infra layer for physical RSI
AI
EnactraAI ranks the open model #12 and says it replaces SWE-2 on BuildingBench’s cost–quality Pareto frontier.
AI
EnactraAI reports an 81.9 overall score for 3D building reconstruction from real photos, up from the previous version’s 72.9.
AI
EnactraAI says it tested the models on BuildingBench, which measures reconstruction of real buildings from photographs.
AI
EnactraAI says WorldBox’s spatial memory management and sandbox help agents remember where resources are and use that knowledge.
AI
EnactraAI says its spatial-reasoning sandbox helps agents manage long-term memory, infer mechanics and simulate reasoning.
Technology
Meta says its 816B-parameter model can run locally across two M5 Ultra Mac Studios.
AI
Google says Argon is available through its paid API tier and to Pro and Ultra subscribers in Gemini and Search.
YouTube
AI
EnactraAI reports a cost of $1.34 per building for Haiku 5.5 (max), versus $15.03 for Sonnet 5.
AI
Enactra says Opus 5.5 made 133% more progress and GPT-6 Astra 50% more after gaining memory of their Minecraft world.
AI
EnactraAI says GPT‑6.1 Sol scored 85.1 at a median cost of about $1.70 per building. Sonnet 5.5 (max) scored 76.9, but cost more per building than Sonnet 5.
AI
EnactraAI puts Cognition’s SWE-2 (max) 2.5 points behind GPT-6 Sol (ultra) and says SWE-2 was free to use as of September 25, 2026.
AI
EnactraAI reports a sizable cost gap: GPT-6 Astra (ultra) costs $7.05 per building versus about $45.47 for Opus 5.5 (max), with scores of 84.3 and 86.8, respectively.
AI
SpaceXAI calls it an improvement over Grok 4.6 at the same price and speed. Artificial Analysis reports better coding results, but more than double the output-token use in a separate evaluation.
AI
EnactraAI says the matchup took place in a New York City simulation Astra created. The world kept moving while the models thought, leading to some awkward collisions with pedestrians and cars.
AI
EnactraAI says its BuildingBench benchmark tests whether coding agents can turn real-world images into coherent 3D buildings, and at what cost.
AI
EnactraAI says GPT-6 Astra achieved the highest score in its comparison at about 80% lower cost than Fable 5.1.
AI
A user describes Grok’s progress from plain white boxes in version 4.2 to rough approximations in 4.5 and complete, detailed buildings in 4.6.
AI
The creators say their interactive model has three visual layers—anatomy, blood vessels and nerves—with 8,100 modeled vascular and neural branches.
AI
EnactraAI posted a video comparing three AI models on spatial reconstructions of buildings.
AI
Reply corrects earlier cost labels with video of three-panel UI for Geisel Library reconstructions.