The article argues that AI token costs have dropped so dramatically that per-token billing and metering are becoming impractical, suggesting a shift toward flat-rate or alternative pricing models. The decline in token costs reflects rapid improvements in model efficiency and infrastructure scaling.
Background
AI inference costs have dropped over 90% in recent years due to model optimization and hardware improvements, prompting industry discussions about whether the dominant per-token pricing model remains viable.
- Source
- Hacker News (RSS)
- Published
- Sep 23, 2026 at 05:21 PM
- Score
- 6.0 / 10