SansCreates
spend
$10,668.74
tokens
16.3B tokens
models
7 models
summary
- requests
- 37,034
- active days
- 76
- per active day
- $140.38
- per request
- $0.288
- tokens per request
- 441K
- biggest day
- $437.60
- first reported
- 2 Jul 2026
- last reported
- today
what they pay for
$100.00/mo
- Anthropicmax 5xsubscription$100.00/mo
$100.00 a month in subscriptions against $5,128.12 of usage in the last 30 days at API rates, so they carried 51.3× their own cost.
Self-reported. Nothing here is checked against a provider.
where the money went
| bucket | tokens | cost | share |
|---|---|---|---|
| cache read | 15.9B | $6,625 | 62.1% |
| cache write | 370.3M | $3,318 | 31.1% |
| output | 32.6M | $724 | 6.8% |
| input | 521.7K | $1.62 | 0.0% |
93.2% of this is cache, not completions. Context being re-read each turn, rather than anything typed or generated.
cache reuse
43.0:1
15.9B tokens read back for every 370.3M written. Writing to cache costs 1.25–2× the input rate; reading costs a tenth. Higher is better.
Not enough people syncing yet to compare against.
this month
$3,735 so far · $194/day · on pace for $5,676
Projection uses the last fortnight, including days with no usage, over 10 remaining days.
the same tokens, priced elsewhere
| model | total | vs yours | effective | in | out | cache read |
|---|---|---|---|---|---|---|
| your mix · 7 models · 16.3B tokens | $10,669 | — | $0.653 | |||
| haiku-4-5 · used | $2,492 | −$8,177 77% | $0.152 | $1.00 | $5.00 | $0.100 |
| gpt-5.2 | $4,532 | −$6,137 58% | $0.277 | $1.75 | $14.00 | $0.175 |
| sonnet-5 · used | $4,984 | −$5,685 53% | $0.305 | $2.00 | $10.00 | $0.200 |
| gemini-3-pro | $5,049 | −$5,620 53% | $0.309 | $2.00 | $12.00 | $0.200 |
| opus-5 · used | $12,460 | +$1,791 17% | $0.762 | $5.00 | $25.00 | $0.500 |
| opus-4-8 · used | $12,460 | +$1,791 17% | $0.762 | $5.00 | $25.00 | $0.500 |
| fable-5-1 · used | $12,964 | +$2,296 22% | $0.793 | $10.00 | $50.00 | $0.250 |
USD per million tokens. Effective applies each rate card to your own mix, the only comparison that holds when cache reads are 62.1% of the bill. Published rates only: a weaker model often needs more turns, and those are not counted.