Overview
- This week major labs moved fast to defend customers by cutting costs and efficiency losses: OpenAI introduced GPT‑6 Sol and GPT‑6 Luna and cut API prices roughly in half, while Anthropic released Claude Opus 5.5 and said it runs about 40% cheaper than the prior Opus model.
- Large customers have concrete targets to shift workloads: AT&T currently runs about 40% of AI tasks on open models and plans to reach roughly 70% within a year, and Match/Tinder and other firms have already re-routed simpler queries to lower-cost models.
- Companies say the appeal of open-weight models is lower per-workload cost and the ability to run models on company servers so sensitive data never leaves their control, a point highlighted by a data-center operator that keeps proprietary data in a private model.
- Adopting open-weight models brings tradeoffs because firms must add hardware, software and ops work to manage models on-premises or in private clouds, which forces CFOs and CTOs to weigh direct cost savings against new infrastructure and security responsibilities.
- The shift is widening beyond tech into finance, logistics and industrial firms and could spur more on-site compute investment, a greater role for non‑proprietary models from global providers, and continued price-pressure that will shape which tasks stay on frontier APIs.