Where is Supersonic Labs headquartered?

Supersonic Labs is headquartered in Brazil. The company has a team of 1-10 people.

What does Supersonic Labs do?

Supersonic Labs operates in the Natural Language Processing sector. Brazilian independent AI research lab building compact, open-weight decision models that run locally on CPU hardware. Compact, open-weight decision models (classification, routing, scoring, yes/no) that run on commodity CPU hardware with no GPU required — bringing capable AI to the hardware people already own.

When was Supersonic Labs founded?

Supersonic Labs was founded in 2026. It is based in Brazil.

Who is the CEO of Supersonic Labs?

João Davi is the CEO of Supersonic Labs. The company was founded in 2026.

Where is Supersonic Labs located?

Supersonic Labs is headquartered in Brazil.

What funding stage is Supersonic Labs at?

Supersonic Labs is currently at the Bootstrapped stage.

How many employees does Supersonic Labs have?

Supersonic Labs has approximately 1-10 employees.

Supersonic Labs

Category: Natural Language Processing

Brazilian independent AI research lab building compact, open-weight decision models that run locally on CPU hardware. Supersonic Labs was founded in 2026. The company is led by João Davi. Based in Brazil. Team size: 1-10. Latest round: Bootstrapped.

Founded: 2026
Headquarters: Brazil
Team size: 1-10

Value proposition

Compact, open-weight decision models (classification, routing, scoring, yes/no) that run on commodity CPU hardware with no GPU required — bringing capable AI to the hardware people already own.

Products and solutions

Julia 1 — 144.3M-parameter open-weight (Apache 2.0) decision model built on JHU CLSP's mmBERT-small encoder with a custom decision head, supports three decision types (choice, score, noul/boolean) via one API, runs on CPU (Python 3.11+) and in-browser via ONNX/WebGPU. Julia-1-ONNX export and a Julia-1 Decision Demo Space on Hugging Face. Julia 2 (own foundation architecture) reportedly in development. Hosted API announced but not yet public.

Unique value

A 144.3M-parameter model that beats proprietary structured-decision rivals on 3 of 4 benchmarks while running on a laptop CPU, trained for ~US$104 in cloud GPU spend.

Target customer

Developers and enterprises needing local, low-cost classification, routing, and structured-decision inference (e.g., support-ticket routing, intent classification, risk scoring) without GPU infrastructure.

Industries served

Developer tools, enterprise software (customer support routing, intent classification), edge/local AI

Technology advantage

Finite-choice decision architecture (no free-text generation) keeps parameter count and latency low; multilingual foundation (mmBERT-small, 1,800+ languages); runs on CPU/phone/browser with ~550 MiB FP32 checkpoint; permissive Apache 2.0 open weights; extremely low training cost (~US$104).

How they differentiate

Unlike frontier chatbot labs, Supersonic Labs deliberately goes small and narrow: a decision model that selects among supplied options rather than generating text, optimized for CPU/edge deployment and open-weight distribution.

Main competitors

TypeSafe AI (Jev, proprietary structured-decision model), Fastino Labs (GLiNER2.5-Decide, 340M open-weight decision model), general small-model/encoder providers

Key partnerships

Builds on JHU CLSP's mmBERT-small encoder (open-source foundation), community Swift/MLX ports (julia.swift, Julia-1-MLX)

Major milestones

Released Julia 1 on September 26, 2026 — a 144.3M-parameter Apache 2.0 decision model, benchmarked Sept 24, 2026 (73.15% typed decisions, beating Jev reference on 3 of 4 tasks), published weights on Hugging Face, announced Julia 2 in development.

Growth metrics

Julia 1 Hugging Face model: ~145 likes, ~274 downloads shortly after release; ~4K X/Twitter followers (@supersonicai); 4 team members listed on Hugging Face org.

Market positioning

Early-stage challenger in the emerging "small/efficient decision model" niche, competing with proprietary structured-decision models (TypeSafe's Jev) and open-weight alternatives (Fastino's GLiNER2.5-Decide).

Geographic focus

Brazil / global open-source AI community

Patents and IP

Open-weight model released under Apache 2.0 license (no public patents disclosed)

About João Davi

Product & Technology Leader, Full-Stack & App Engineer (8+ years); ex-Head of Product at Wplace (collaborative platform).

Latest news about Supersonic Labs

Supersonic Labs Releases Julia 1, a CPU-Ready Open Decision Model

More Natural Language Processing companies

Official website: https://supersoniclabs.ia.br/