Supersonic Labs Releases Julia 1: A 144.3M-Parameter Open Decision Model That Runs on a CPU
144.3M parameters, CPU-native, Apache 2.0: Supersonic Labs drops Julia 1, a lightweight decision model that beats some benchmarks but trails on Banking77.
Why it matters
A small open-weight model optimized for discrete choice tasks (ranking options given context) challenges the assumption that useful models require massive scale or GPU resources. The mixed benchmark results and narrow use case limit broad impact, but CPU deployability matters for edge and cost-constrained deployments.
The key facts
15 to knowJulia 1: 144.3M parameters, built on mmBERT-small
Task: takes context, question, 2–20 options; returns one choice with probabilities
Runs on CPU; Apache 2.0 license
Beat Jev reference values on 3 of 4 pilots
Trailed on Banking77 (72-label test)
Released by Supersonic Labs
Model size: 144.3M parameters
Built on mmBERT-small foundation
Runs on CPU (no GPU required)
Input format: context + question + 2-20 options → single choice with probabilities
License: Apache 2.0 (open-weight)
Performance: beat Jev reference values on 3 of 4 pilots
Failed on Banking77 test (72-label classification benchmark)
No pricing, deployment limits, or production deployment data disclosed
Source: MarkTechPost summary of vendor release; no independent testing cited
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Supersonic Labs has released Julia 1, a 144.3M-parameter decision model built on mmBERT-small. It takes context, a question, and 2 to 20 options, then returns one choice with probabilities. The model runs on a CPU and ships under Apache 2.0. It beat Jev reference values on 3 of 4 pilots but trailed…