SKY-001 claims CAD/BIM benchmark win over GPT-6 Astra

Sam Pasupalak @spisallyouneedOriginally posted 2026.10.20

Open original on X

1/n The benchmark results back up what we have argued from day one: LLMs are autoregressive by nature, which means they predict one token at a time, but they were never built for visual reasoning and long-horizon tasks. We tested SKY-001 against the frontier models, GPT-6 Astra, Claude Opus 5.5, etc, in both CAD and BIM. In both cases, our model outperformed the LLMs, setting a new standard for AI efficiency.