Models

SZkrO model line

Current releases are still in development. The R0 phase selected the best practical base for this hardware and froze a pre-training benchmark.

Development track

SZkrO 1.0 · 3B

Planned first tuned SZkrO release line. It is being developed from Granite 4.1 3B, with R1 intended to add SZkrO-specific training rather than simply rebrand the base model.

FoundationGranite 4.1 3B
Ollama tag in R0granite4.1:3b-q4_K_M
Current statePre-training baseline
Fine-tuningNot started in R0
TargetLocal practical AI
Transparency note: the current SZkrO development line is based on Granite 4.1 3B. A future “SZkrO Native” line is a separate long-term goal and would mean a model trained from scratch rather than inherited weights.