Vals analysis: open-weight models performing multi-stage tasks, like building a web app, can have an environmental impact 10K times greater than simple queries (Bloomberg)

OpenAI launched GPT-6 Astra, trained on a record-size run with over 100,000 GPUs at its Texas Stargate facility and priced at $10 per million input tokens and $50 per million output tokens. In the ARC-AGI-3 benchmark it scored 62.7% with the standard harness ($26 K) and 99.9% with a provider-adapter harness ($19 K), surpassing Claude Opus 5 (30.2%) and GPT-5.6 Sol (7.8%). Vals analysis notes that open-weight models executing multi-stage tasks, such as building a web app, can require up to 10,000 × the energy of simple query workloads.