Skip to content

Latest commit

 

History

History
13 lines (9 loc) · 730 Bytes

File metadata and controls

13 lines (9 loc) · 730 Bytes

TT-4. "Why now: inference costs fell ~1000x in three years"

Punchline

"GPT-3 was ~$60/M tokens in 2021. GPT-4 launched at ~$30/M in 2023. GPT-4-class quality at ~$0.30/M by 2024. Today, frontier models like Claude Haiku, Gemini Flash, GPT-5 mini hit pennies per million. Three years ago, an agent that called the model 50 times in a loop was economically unviable. Today it costs a few pennies."

Why this lands

Answers the "why now" question with a chart, not vibes.

NeuralSeek tie-in

This 1000x deflation is the unlock. But it makes the governance problem 1000x bigger too — every call is a new chance to hallucinate, leak, or run away on tokens.

Use

Pair with any "AI is finally ready for enterprise" pitch.