Relantic

We ran @Elorian's new Glimpse model on RealityBench's physical work tasks: it reads labels as well as Argon, but has gaps in fine-grained visual reasoning: tape measures, lag bolts. Also, reasoning adds up quickly: our run cost $18 for a score that Gemma 4 31B matches for $0.70.

Frontier VLMs have a reality gap: Of 30 models, they now match a human rater at judging physical work, but read a tape measure right only 36% of the time.

We need a benchmark for reality. Introducing RealityBench: 3D worlds you can bend 🧵

Reality is Messy. We're here to help. relantic.com