jiwo-4b
A Qwen3.5-4B decision model by Jaime Ferrando Huertas with a trained readout: a state and named typed questions (choice up to 255 options, score on 2 to 10 levels, noul) go in, and one forward pass returns a probability for every option, with no generated text.
Full weights plus a readout file (8.4 GB). Trained with an 8,192-token context; mostly English. jiwo serve runs a Jev-style server on /v1/systemone; the card says TypeSafe does not endorse it. The maker's own run of the third-party Decision Index 0.2.1 gives 44.97 for the 4b and 28.00 for the 0.8b, not yet verified by the board. Sibling jiwo-0.8b (0.75B). Code is MIT.
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one probability
At a glance
| Parameters | 4.2B |
| Base model | Qwen/Qwen3.5-4B |
| Maker | Jaime Ferrando Huertas |
| Released | 2026-10-05 |
| License | apache-2.0 |
| Reported latency | 53 ms median per request on one H100 (48 ms for jiwo-0.8b) |
Get the weights
pip install systemonemodels
systemone pull jaime-ferrando-huertas/jiwo
The files are served from the maker's Hugging Face repository, eljiwo/jiwo-4b, and verified against the checksums recorded here.
Read more
This page was opened by System One for Jaime Ferrando Huertas, who can claim the organisation and take it over at any time.