Supersonic Labs releases Julia 1, a 144.3M-parameter open decision model
Supersonic Labs, a Brazilian AI lab, has released Julia 1, a compact decision model with 144.3M parameters. Unlike standard chatbots, Julia 1 does not generate text; instead, it accepts context, a question, and 2 to 20 candidate answers to select one option and return probability scores. The model is built on JHU CLSP’s mmBERT-small encoder and is available on Hugging Face under the Apache 2.0…
Key points
- Julia 1 is a 144.3M-parameter model released under Apache 2.0 license by Supersonic Labs.
- The model runs on CPUs with a median latency of 33.15 ms on Apple M4 hardware.
- Training cost approximately US$104.08 in cloud GPUs, with a planned API price of $0.025/MTok.
The lab reports that Julia 1 outperformed TypeSafe’s Jev reference values in three of four benchmark pilots, including AG News and DAIR Emotion. However, it significantly underperformed on the Banking77 dataset, scoring 64/100 compared to Jev’s 87/100 reference. Total cloud GPU training costs were approximately US$104.08. On-device latency tests showed a median of 33.15 ms per decision on an Apple M4 chip, while an Intel Core i5-1235U took 107.83 ms for simpler tasks.
Supersonic Labs notes that Julia 1 is limited to comparing supplied options and cannot handle multi-step calculations or missing facts. A hosted API is planned at $0.025 per million tokens but is not yet open. Julia 2, featuring a proprietary foundation architecture, is currently in development.
Model pages: Julia 1 → · Jev →
Supersonic Labs Releases Julia 1: A 144.3M-Parameter Open Decision Model That Runs on a CPU
MarkTechPost · 26 September 2026
Loading the full article…
This text was published by MarkTechPost and written by Michal Sutter. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Claude Opus 5.5 tops benchmark over OpenAI’s Astra and Fable 5.1 · 3 src
- Anthropic releases Claude Opus 5.5 with 40% lower costs · 15 src
- OpenAI releases GPT-6 Sol and Luna for cheaper coding and high-volume tasks · 5 src
- OpenAI releases GPT-6 Sol with 7.7% improvement and half API cost · 6 src
- Meta's Muse agent gives every user a full cloud computer running Ubuntu Linux · 7 src
Comments
via GitHub Discussions