LUA VISIONLUA Vision official launch at MinervaHUB
Invitation-only event in São Paulo, from 4 pm to 9:30 pm.
For anyone writing about LUA.
Published coverage, official facts and team contact. Every figure quoted here comes from the public registry, with date and method.
LUA news.
A run of 49 high-risk cases
Health, law, finance and everyday life in Brazil, with failure criteria written before the test.
LUA Vision at Gramado Summit
Appearance covered by Exame.
Genesys PI family
LUA's third model generation, in three deployment formats.
LUA in the press
ExameMay 6, 2026
At Gramado Summit, AI dominates debates, with a warning: people stay at the center
Event coverage presenting LUA Vision as a startup with its own model trained for the Brazilian context in tax, law and business.
NeoFeedApr 30, 2026
Feature on language models built in Brazil
Introduces the founders, the origin of the models in Paulo Câmara's doctoral thesis and the company's pilot projects.
For your fact box.
Legal name
LUA Vision Tecnologia Ltda.
Headquarters
São Paulo, Brazil
Models
Genesys PI family
Open →Products
8th.vision
Open →Spelling
LUA Vision, with LUA in capitals.
Logo
LUA logo with transparent background, for editorial use.
Open →Notes
- 012025Founded
The company started operating in late 2025, in São Paulo, after three years of doctoral research.
- 02about 20Pilot projects
Figure given by the founders, in segments such as health and education.
- 0398.2%LiveBench, global average
Evaluation run by LUA with the LiveBench evaluation code, on the public questions of the 2026-01-08 set: 682 questions across five categories (data analysis, instruction following, language, math and reasoning), at temperature zero. Coding was left out because the set has no public questions in that category. Model evaluated: LUA Genesys, second generation.
- 0449 of 49High-risk cases with no unsafe answer
49 high-risk questions across six areas: health, finance, facts, law, attempts to bypass rules and everyday life in Brazil. Each answer is checked against what must appear and what must not, defined before the test. Result: 46 answers passed, 3 questions stopped before reaching the model and no failures.