Julia 1
#11 in Small & Edge ModelsSupersonic Labs · v1 · 3× · last seen Sep 28, 2026
48
Momentum
Julia 1 is a compact decision model with 144.3 million parameters released by Supersonic Labs (Brazil), based on the multilingual encoder mmBERT-small. Rather than generating text, it processes context, a question, and 2 to 20 possible answer options, returning probability values for classification, routing, ranked assessments, and yes/no decisions. The model runs locally on CPU or BF16-capable GPU as well as as an ONNX build in the browser via WebGPU and is Open Source under Apache 2.0; a hosted API is planned but not yet publicly available.
Momentum trend
30.06.28.09.
Features
| Key Benchmark (%) | Typed Decisions: 73.15% (1,463/2,000) vs. Jev reference 72.70% |
| Context Window (Tokens) | Up to 8,192 combined tokens (benchmarks measured with a 1,024-token limit) |
| License | Apache 2.0 (open weights) |
| Multimodality | No multimodal support; text-only decision model (multilingual, no image/audio) |
| Platform | Runs locally via Python 3.11+ on CPU or BF16-capable GPU; ONNX build in-browser via WebGPU |
| Price per 1M Tokens | $0.025 per million input tokens planned (API in development, output free) |
| Release Date | September 26, 2026 |