Across anonymized, opt-in MyTokenTracker contributor data, developers spent $9843.66 over 114 days for 194,648,481,236 tokens. That works out to about 19,773,995 tokens per dollar, with cached reads included, which is a very low effective cost per token.
The big takeaway is that real-world AI spend is being stretched by volume and reuse, not just raw model pricing. For developers, the budget story is less about isolated expensive calls and more about how far a dollar goes when traffic is large and cached reads are part of the mix.
This dataset is useful because it reflects actual community usage, not a benchmark or a vendor price sheet. Over 114 days, the tracked spend stayed at $9843.66 while token volume reached 194,648,481,236, which suggests that many workflows are operating at very high throughput without proportionally high spend.
About 19,773,995 tokens per dollar is the clearest signal here. That figure implies that token efficiency matters a lot more than headline model cost when cached reads are included, because repeated or reusable prompts can dramatically dilute spend across a large token base.
For budget planning, this points to a practical reality: the expensive part is not always the token count itself, it is the uncached, non-reusable work. Teams that can maximize cache hits and keep prompts stable will usually get much more output from the same budget than teams that constantly generate fresh context.
- Tracked community spend was $9843.66 over 114 days, so the budget footprint is real but not extreme.
- 194,648,481,236 tokens against that spend shows that scale can be cheap when reuse is high.
- About 19,773,995 tokens per dollar means cached reads materially improve effective efficiency.
- Budget conversations should focus on cacheability and prompt reuse, not just model choice.
About this data. Figures are from MyTokenTracker live data (community dashboard, open data API), free to use and cite under CC BY 4.0. Snapshot: October 22, 2026.