AI token prices

Is AI getting cheaper to use?

Contributes to: Can AI do the work?. Signal score: 47 / 100. Relative weight within this question: 1.

StatusGetting cheaper
Latest value$2.00
Year over year-27%
Distance from high-88%

Observed trend

Median listed price in dollars per 1M tokens, per month, with 3 input tokens per output token.

Median listed price in dollars per 1M tokens, per month, with 3 input tokens per output token. · 2023-10 – 2026-11

2026-11 — Median listed price in dollars per 1M tokens, per month, with 3 input tokens per output token.: 2
View data table
Period / observationMedian listed price in dollars per 1M tokens, per month, with 3 input tokens per output token.
2023-1016.44
2023-1116.44
2023-1212
2024-014.1
2024-0212
2024-034.1
2024-044.05
2024-054.05
2024-063.25
2024-073.75
2024-083
2024-093.25
2024-103.75
2024-113.25
2024-123.25
2025-013.75
2025-023.25
2025-033.75
2025-043.25
2025-053.25
2025-063
2025-073
2025-083
2025-093
2025-102.89
2025-112.75
2025-122.75
2026-012.75
2026-022.75
2026-032.75
2026-041.93
2026-052
2026-062
2026-072
2026-082.75
2026-091.56
2026-101.93
2026-112

What it measures

Metric
Median listed price per 1M tokens across AI models from their makers, with 3 input tokens per output token
Solved if
Using AI becomes close to free
Signal score
47 / 100. Down 88% since October 2023, when the data starts. How it is scored
Status rule
Getting cheaper when the median listed price is lower than a year before.

Are prices falling per model?

Mostly not. The median falls because newer, cheaper models arrive, not because existing models get repriced.

Post by Dax Raad: with current token prices i wouldn't really say "cost of software is going to zero". our token cost is double our payroll which means even though we're 25 it's like we're 75. and i feel like our output is similar to a team of 75 pre-ai. we're just able to do it without all the downsides of a larger team. this is still a good thing. we couldn't hire 50 more people. so much overhead and impossible to retain a high bar
Dax Raad, co-founder of OpenCode. One team's experience, not a measurement.

Update history

The reading on each date we checked, separate from the historical observations above.

DateValueYear over yearStatus
2026-11-02$2.00-27%Getting cheaper
2026-11-01$2.00-27%Getting cheaper

Sources

SourceRoleMeasuresLatest data
LiteLLM model pricesScored measurementListed prices for AI models over time2026-11
OpenCode DataSupporting contextAverage AI usage cost per session, quoted by handNov 1, 2026