Research
Every number on this site can be rerun from what we publish.
Oct 5, 2026OpenDecision technical reportFast schema-conditioned decisions with packed candidate encoding. (arXiv version pending)ReportPDFOct 5, 2026OpenDecision modelsLarge, Large-Packed and Small, Apache-2.0.ModelsHugging FaceOct 5, 2026OpenDecision predictionsModel predictions behind the reported numbers.DatasetHugging FaceOct 5, 2026OpenDecision codeTraining, evaluation and inference.CodeGitHub
OpenDecision models
GPU-seconds per 1,000 decisions on one NVIDIA L4. Price per 1M tokens on the Tokz API.
ModelParamsSuite macro-F1GPU-s / 1kAPI priceLicense
OpenDecision-Large
Params434M
Suite macro-F187.9
GPU-s / 1k77.4
API price$0.035 / 1M
LicenseApache-2.0
OpenDecision-Large-PackedDefault
Params434M
Suite macro-F187.2
GPU-s / 1k69.1
API price$0.025 / 1M
LicenseApache-2.0
OpenDecision-Small
Params71M
Suite macro-F179.3
GPU-s / 1k17.5
API price$0.025 / 1M
LicenseApache-2.0
OpenDecision-Large leads the open models we measured
Average macro-F1 on five classification benchmarks, full test sets, same harness.
OpenDecision / Classification suite
ModelMacro-F1
- OpenDecision-Large87.9
- OpenDecision-Large-Packed87.2
- SemIf (4B LLM)86.3
- JevK5 (4B LLM)86.3
- Laya81.5
- GLiFormer80.5
- OpenDecision-Small79.3
- GLiNER2.574.4
Cite the report
OpenDecision.bib
@misc{tokz2026opendecision,
title={OpenDecision: Fast Schema-Conditioned Decisions with Packed Candidate Encoding},
author={Tokz Labs},
year={2026},
note={Technical report}
}