[Daily Picks]

Press coverage (e.g. TechRadar) says Revolut moved routine coding off Claude to self-hosted open-weight models and claims roughly an order-of-magnitude inference cost cut—while still noting GPUs, power, and ops are real expenses. In local-community terms, it is another data point for when self-host wins (repetitive coding, privacy, flat capacity) versus when frontier APIs still earn their keep on hard, long-horizon tasks.

Practical tip: DIY teams should cost hardware + electricity + babysitting hours, not only "$/M tokens." Self-host shines for high-volume, similar tickets; keep a frontier API for the nasty 10%.

Where do you draw the line between local/self-hosted coding models and paid frontier APIs?

Source: https://www.techradar.com/pro/we-were-highly-skeptical-about-the-cost-of-ai-revolut-reveals-it-cut-ai-coding-costs-by-cutting-out-claude-entirely