MODELS
Model selection
A short method for choosing between Z1, Z1 Code, H1 and Project ZERO Alpha.
Tessen does not rank its families against other vendors' models, and no benchmark result is published here. Choose by the kind of work, then check on your own data.
| If your task is | Start with | Why |
|---|---|---|
| Chat, drafting, summarising, short extraction | z1 | Lowest cost per token, fast. |
| Writing, reviewing or fixing code | z1-code | Priced and tuned for programming work. |
| Multi-step reasoning where a wrong answer is costly and you can check it | h1 | It reasons before answering. Slower, and reasoning tokens are billed as output. |
| Coding in OpenCode, free, within a daily allowance | project-zero | Free for enrolled developers; never touches your wallet. |
Evaluate on your own task
- Collect 20 to 50 real inputs
Use the cases you actually need handled, including the awkward ones.
- Run them on two models
Send the same messages to
z1and the candidate with the samemax_tokens. - Compare answers and cost
The final streamed frame reports
cost_usdper request, so cost per task is measurable. - Pick the cheapest model that passes
Move up a family only where the cheaper one fails.
Note
Send an explicit model id in every request. A model your account may not call is refused; it is never swapped for another.