factual
Serving cost falls fast through algorithmic gains
The cost to serve a given model quality falls rapidly through algorithmic improvement rather than bigger models, with GPT-3-level serving about 2,000 times cheaper now and DeepSeek-era and open-source models cheaper still.
factualpending
Speaker
Dylan PatelEvidence Quote
“the cost to serve a model quality of GPT-3 is like 2,000 times cheaper now. People freaked out about DeepSeek because it was 500–600 times cheaper.”
Created: 11/16/2026, 4:01:45 AM
Your notes
Loading notes...