Models

Supersonic Labs releases Julia-1, a 144.3M-parameter decision model with a 550.5 MiB checkpoint

2026-09-27 · that day's edition · one of the five

Built on mmBERT-small, the Apache-2.0-licensed classifier takes a state, a question and up to 20 options, and returns one of them.

huggingface.co, quoted / SupersonicLabs/Julia-1 · source · Quoted under the quotation exception; no licence granted

Supersonic Labs released Julia-1, a 144.3-million-parameter open-weights model for classification and binary judgments rather than text generation. Built on JHU CLSPs mmBERT-small, it accepts a context, a question, and 2 to 20 candidate answers, and its FP32 weights occupy 550.5 MiB. The model artifacts are licensed under Apache 2.0, and the checkpoint evaluation, dated September 24, 2026, is the first release in what the company calls the Julia family of decision models.

What this rests on

Julia-1 has 144.3 million parameters and is built on JHU CLSPs mmBERT-small.

SupersonicLabs/Julia-1 · Hugging Face (Supersonic Labs model card) · 2026-09-24
The model artifacts are licensed under Apache 2.0
The model weighs about 550.5 MiB and is licensed under Apache 2.0.

SupersonicLabs/Julia-1 · Hugging Face (Supersonic Labs model card) · 2026-09-24
The model artifacts are licensed under Apache 2.0
The model accepts a state, a question, and 2 to 20 candidate options to produce a decision.

SupersonicLabs/Julia-1 · Hugging Face (Supersonic Labs model card) · 2026-09-24
The model artifacts are licensed under Apache 2.0

We checked every sentence above against its source by opening it. Nothing appears on this site that we have not opened and linked.

Filed under

Also that day

Agents and toolingClaude Code 2.1.280 sets Opus 5.5 as the default Opus model, with a 1M-token window
Research worth knowingDeepSeek describes DSec, a sandbox platform for agentic training handling 3 million sandboxes a day
Research worth knowingPaper describes PrePssmCas, a classifier that fuses language-model and PSSM features for Cas proteins