OpenAI is pitching GPT-6.2 Astra as a cheaper version of Claude Fable 5.2. For construction I don't think cost matters much. If I'm modelling a $5 million warehouse, I care whether the model is right, not whether it cost $5 or $20 in tokens. I gave Astra a set of 2D structural drawings for a warehouse and one prompt: build the 3D model, write the construction programme, then combine them into a 4D simulation. The 3D model looked more detailed than Fable's, but Fable followed the drawings more closely. Astra added cladding that wasn't on the drawings and missed some of the footings. For take-offs and scheduling I'd still use Fable. The programme logic was sound (weather, crane lifts, subcontractor calendars), but the activity names and groupings were hard to follow, and it ignored laydown and site access. Astra did better on an open-ended brief: run 5,000 simulations and find the fastest way to finish. Most construction work needs clear instructions, so that's a narrow use case. If you're already set up in ChatGPT, I wouldn't switch just for this. Free resource: ContractorOS construction AI benchmarks (https://contractoros.build/benchmarks/). Free to view, and shows how each model scores on reading real construction drawings, on its own and inside apps like the Claude desktop app. What are you using Astra for, if anything? https://www.youtube.com/watch?v=Br7SzhV4Qd0