Trends
How the cost of AI moves over time. Every chart is free to embed and download, with its sources named.
What a strong model costs, over time
Cheapest blended price, per 1M tokens
- 70% on GPQA Diamond
- At or above GPT-4 on LMArena
Solid: cheapest model scoring 70% or more on GPQA Diamond (Epoch AI). Dashed: cheapest model rated at or above GPT-4 (June 2023) on the LMArena text leaderboard, a lower bar that small models such as Llama 3.1 8B clear. Blended: three input tokens per output token. Sources: LiteLLM price history (MIT), OpenRouter, Epoch AI (CC BY 4.0), LMArena (CC BY 4.0).
The price of a hard-science pass mark
Cheapest blended price, per 1M tokens
- 50% on GPQA Diamond
- 70% on GPQA Diamond
Cheapest model at or above each GPQA Diamond score (Epoch AI). Blended: three input tokens per output token. Sources: LiteLLM price history (MIT), OpenRouter, Epoch AI (CC BY 4.0).
China and the US at the same quality
Cheapest blended price, per 1M tokens
- China
- United States
Cheapest model from each developer country scoring 60% or more on GPQA Diamond; at least three models per country. Sources: LiteLLM price history (MIT), OpenRouter, Epoch AI (CC BY 4.0).
Open against closed weights
Cheapest blended price, per 1M tokens
- Open weights
- Closed
Cheapest model scoring 60% or more on GPQA Diamond, by whether its weights are published. Sources: LiteLLM price history (MIT), OpenRouter, Epoch AI (CC BY 4.0).
List-price changes per month
Changes per month
- Down
- Up
Every list-price change across models, hosts and units, by direction. Sources: LiteLLM price history (MIT), OpenRouter.
Same model, different host
Dearest ÷ cheapest host, median
For each model with three or more hosts, the dearest over the cheapest input price; the median across models. Sources: OpenRouter.
What a top-10 image costs
Per image
- Median
- Lowest
Per-image price at a standard size among the ten best-rated LMArena text-to-image models. Sources: LiteLLM price history (MIT), LMArena (CC BY 4.0).
What a second of video costs
Per second of video
- 1080p, lowest
- 1080p, median
- 720p, lowest
Per-second list price across video models at each resolution. Sources: LiteLLM price history (MIT).
The GPU rent index
Median on-demand price, per GPU-month
- B200
- H200
- H100
Median on-demand price per GPU-hour across every cloud and marketplace offer × 730 hours. Series starts on 22 October 2026. Sources: Cloudadapt GPU prices (CC BY 4.0).
Quality data: LMArena (CC BY 4.0), Epoch AI (CC BY 4.0). Prices: OpenRouter, LiteLLM (MIT), provider pages. List prices in US dollars before discounts and taxes; other currencies converted at the European Central Bank reference rate. The figures on these pages are free to reuse under CC BY 4.0 with a link to this page.