Experiments
EXPERIMENT 005

LLM Price Volatility

Tracking the exponential price decay of frontier AI model APIs — dollar per million tokens, from GPT-3 to GPT-4.1 nano.

$0.10$0.50$1$5$10$50$100$/1M tokens (input)20212022202320242025
OpenAI
Anthropic
Google
Meta (hosted)
Updated 2026-03-18 · 35 price points tracked

What You're Seeing

This chart tracks the input price per million tokens for frontier AI models from 2021 to present. The Y-axis uses a logarithmic scale — each gridline represents a 10x price difference.

Two patterns emerge: exponential decay as compute costs fall, and sudden step drops when new model generations launch. GPT-3 Davinci cost $20/1M tokens in 2021. GPT-4.1 nano costs $0.10 in 2025 — a 200x reduction in four years.

The chart updates weekly. Data sourced from official provider pricing pages, launch announcements, and llm-prices.com.

Methodology

Prices shown are list API prices (no volume discounts, no cached-input pricing). Input pricing is the default view — output pricing is typically 2-4x higher. All prices are USD per million tokens.

For open-weight models (Llama), we track hosted API pricing via Together AI as a representative provider. Actual costs vary by provider.

Share

Ready to participate?

Subscribe to Nothing