# A Thermal Budget, Not Model Size, Decides What Runs On-Device

One preprint routes queries across device, edge and cloud tiers with a reinforcement-learned policy that treats heat as the binding constraint, while another searches architectures that fit inside microcontroller memory.

- Published: 2026-09-28T06:48:56.769Z
- Canonical: https://polylog.news/ai/2026-09-28/a-thermal-budget-not-model-size-decides-what-runs-on-device
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.LG](https://arxiv.org/abs/2609.30270), [arXiv cs.LG](https://arxiv.org/abs/2609.30272</url)

The case for on-device inference is usually argued on privacy and cost. A new preprint, HybridInfer, starts from the same premise, that the local tier is preferred when adequate because data stays on the device, it works offline and it incu…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-09-28/a-thermal-budget-not-model-size-decides-what-runs-on-device (subscription information: https://polylog.news/pricing).