What this costs against going direct
The comparison the landing page makes, set out in full. Every figure is a published rate rather than an estimate, and the date they were checked is below the table.
Per million tokens
| MODEL | INPUT | OUTPUT | DIFFERENCE | ||
|---|---|---|---|---|---|
| REFERENCE RATE | AI FROM U SCALE, EST. | REFERENCE RATE | AI FROM U SCALE, EST. | ||
| astra | $10.00 | $0.30 | $50.00 | $1.5 | ≈ −97.0% |
| sol-6.1 | $2.00 | $0.060 | $10.00 | $0.30 | ≈ −97.0% |
| terra | $1.00 | $0.030 | $5.00 | $0.15 | ≈ −97.0% |
Reference prices checked on 27 October 2026.
What this comparison does not say
Provider prices change without notice, and a rate checked on the date above may have moved since. We are an independent service, not affiliated with any model provider, and we do not claim to cost less for every workload — the difference column is the saving on uncached input, and the gap is narrowest on requests that hit the cache, because a cached token here counts half of its model's input rate.