📊 Benchmark — measured, not marketing
Same job. A fraction of the cost.
We ran identical, oracle-graded coding tasks through Depaza and a leading US frontier coding agent. Same passing result — at roughly a third of the cost. Every number below is measured live, not estimated.
Live run · 4 June 2026 · oracle-graded · each provider's own usage accounting
Live head-to-head
Three real coding tasks — a bug fix, a new feature that must preserve an existing API, and a cross-file change. Each agent edited files in an isolated workspace; a hidden executable test graded the result. Cost is each provider's own usage accounting.
| Task | Depaza (core) | Leading US frontier API |
|---|---|---|
| Fix an inverted operator | €0.0057 | $0.0521 |
| Add a utility, keep the existing API | €0.0163 | $0.0657 |
| Wire a config value across modules | €0.0291 | $0.0543 |
| Total | 3/3 · €0.0511 | 3/3 · $0.1721 |
Identical correctness. ~3.4× lower cost. (€ and $ are near parity at current FX, so the ratio reads directly.)
What the same work costs
Total provider cost across the three tasks above — lower is better.
Tier-for-tier list price
Published per-token list prices, blended at a realistic 3:1 input-to-output ratio. Each Depaza tier shown against the comparable tier of a leading US frontier API.
our default model vs the comparable mid-tier
our top model vs the comparable flagship
the expensive half of most bills, default tier
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Depaza Lite | €0.56 | €0.56 |
| Depaza Core | €1.64 | €2.19 |
| Depaza Max | €5.56 | €16.67 |
| BOSS (Unrestricted) | €1.64 | €2.19 |
Prices above are the live, current Depaza API rates — the same numbers we bill.
How we measured it — and where we don't overclaim
- Oracle-graded: each task passes or fails an automated test, not a human judgment.
- Real tokens, real money: cost comes from each provider's own usage accounting, on live accounts.
- Caching caveat: the frontier agent benefits from prompt caching; our EU provider has none — and Depaza is still ~3.4× cheaper.
- A small, honest suite: three tasks show parity on correctness with a large cost gap. We don't claim superiority on every task — harder suites are next.
- Reproducible: same tasks, same oracle, run on demand.
Build on EU First AI — for a third of the price
OpenAI-compatible, prepaid, GDPR-ready. Top up €25, generate a key, and pay only for what you use.