Replace LLM calls with small models you own.
Upload your file. Get back a model that picks one of your options.
Read: reading your file
Pick a text. Get one answer.
An LLM writes a paragraph in seconds. Your model returns one option and a probability for each.
URGENT! You have won a 1 week FREE membership in our 100,000 prize jackpot! Txt the word CLAIM to 81010 now
spam-filter, the model tuned in the run above. Recorded from real calls.
$2, only if it wins.
Scored on rows you labeled yourself, because a win on AI labels held up on human labels only 5 times in 9.
No file: no charge. The run above was scored on AI labels, so it cost nothing.
Measured in the open.
Average macro-F1, five classification benchmarks, open models only
On label sets they weren't trained on, JevK5, SemIf and Laya score higher.
Fine Tune on public datasets, macro-F1, base to tuned
Prices.
| Fine tune with a file | $2.00 | Only if it beats OpenDecision-Small on rows you labeled. |
|---|---|---|
| Fine tune with no file | $0 | No charge. |
| Hosted calls | $0.025 | Per 1M billed tokens, at least 256 tokens a request. About $6.40 per million short calls. |
| Local calls and downloads | $0 | Free. |
| Claude Haiku 5.5, for comparison | about $28 | A million 250-token calls. |
Yours to run.
Call it by name through the Tokz API, or download it and run it free on your machine.
curl https://api.tokz.si/v1/decide \ -H "Authorization: Bearer $TOKZ_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "spam-filter", "inputs": "Hey, are we still on for dinner tonight? I can pick you up around 7"}' # response, trimmed{"model": "spam-filter", "results": [{"decision": {"best": "not spam", "policy": {"spam": 0.000016, "not spam": 0.999984}}}]}