Plummeting Inference Costs and the Multi-Agent Future

bodonoghue85 · x · 2026-07-11

Citing a Stanford report, the author notes that the inference cost for GPT-3.5-level models has dropped by a factor of 280 in less than two years.

The author argues that as inference becomes cheaper, a phenomenon similar to the Jevons Paradox will emerge: lower costs will drive even greater demand, making multi-agent systems more prevalent. The post also mentions that if an open-source model at the "Opus 4.8 level" emerges at a quarter of the current price, many existing forecasts will need to be recalculated.

Original post →

More from AGI Musings

AGI Musings channel →