Press

For the record.

Company facts, current releases, research records, leadership information, and media assets for SimpleDirect® and Vinci.

Press contact Download brand assets Explore current releases

SimpleDirect at a glance

SimpleDirect®: Independent Canadian AI lab operated by Alpine Pacific Trading Inc.
Founded: 2019
Current headquarters: Toronto, Ontario, Canada
Current focus: AI models, research, verification systems, and software developed under Vinci.
Vinci: The name for SimpleDirect’s AI research, models, technologies, and products.
Founder: George Pu, Founder & CEO

Company history

2019 — Waterloo
SimpleDirect founded as a financial technology company.
2025
The fintech business was sunset.
June 2026 — Toronto
SimpleDirect became an AI lab and began publishing its current AI work under Vinci.

Recent releases

Artifact released · September 28, 2026

Vinci MLE 1.0

Open-weight research models for ML-engineering workflows. Availability and licences follow the individual model records below.

Availability and licensing
Model / artifact	Licence	Model card / evidence
Vinci MLE 8B 1.0Public weights	Apache-2.0	Model card
Vinci MLE 30B 1.0Public weights	Apache-2.0	Model card
Vinci MLE 123B 1.0Public weights	Mistral Modified MIT	Model card

Evaluation status

A valid held-out MLE-task evaluation was not completed for 1.0. These releases do not establish proven autonomous ML engineering.

Released weights do not include a complete ML-engineering harness.

Newsroom Technical reports Media assets Press contact

Reporter release package →

Artifact released · September 22, 2026

Vinci Cyber 1.0

Defensive cybersecurity specialist models, fine-tuned from open foundation models. Each size has its own licence and evidence.

Availability and licensing
Model / artifact	Licence	Model card / evidence
Vinci Cyber 8B 1.0Public weights	Apache-2.0	Model card
Vinci Cyber 30B 1.0Public weights	Apache-2.0	Model card
Vinci Cyber 123B 1.0Public weights	Modified MIT (Mistral AI)	Model card

Evaluation status

Evaluation evidence differs by size. Consult each individual model card; the linked 123B aggregate evaluation applies to 123B only, and is not evidence for 8B or 30B.

Do not generalize findings between model sizes or treat local software guards as independent security certification.

What the Vinci Cyber 123B 1.0 evaluation measured

The internal panel contained 24 pairs of flawed and already-correct infrastructure configurations, compared with the parent model mistralai/Devstral-2-123B-Instruct-2512.

The model left 24/24 already-correct inputs unchanged, versus 0/24 for the parent. Targeted scanner findings were cleared without new detected findings in 16/24 flawed inputs, versus 17/24 for the parent: improved restraint did not improve the repair count.

These are BF16-parent-plus-adapter measurements on a small AI-generated panel, without an independent case audit or a run-to-run variance estimate. They do not establish secure or deployable repairs, merged-weight or GGUF results, or results for 8B and 30B.

Inspect the pinned 123B evaluation record

Newsroom Technical reports Media assets Press contact

Supporting technical links

Vinci Cyber 123B 1.0

Reporter release package →

Artifact released · September 6, 2026

Vinci Code CLI

Open-source terminal coding agent. Source availability, the published GitHub artifact and installer version are separate records.

Source licence: MIT.

Published GitHub artifact: 0.0.53. This is separate from the installer channel.

Derived from Pi, by Mario Zechner; preserve upstream attribution.

Evaluation status

A published software artifact is not an independent evaluation or a guarantee of correctness. Review the repository documentation and limitations.

The existing screenshot shows an authentication interface, not a benchmark result.

Newsroom Source repository Technical reports Media assets Press contact

Supporting technical links

Vinci Code CLI

Reporter release package →

Definitive model catalogue

Technical reports

The numbered Vinci technical reports preserve methods, evaluations, corrections and negative findings alongside current work.

September 1, 2026 · Vinci Technical Report No. 3

Runtime Pass Is Not Correctness

Version 1.0 · Draft manuscript

The reasoning-efficiency post-training recipe missed its targets. An evaluator audit found that the original test bank could not support its correctness claims.

A negative result for the recipe and evaluation setup studied in this report; it does not establish that all reasoning-efficiency approaches fail. The manuscript retains its draft status.

Read the manuscript (PDF)DOI: 10.5281/zenodo.22241477

Technical reports →

What we can say on the record

SimpleDirect® is an independent Canadian AI lab operated by Alpine Pacific Trading Inc. in Toronto, Ontario, Canada. The company was founded in 2019, sunset its earlier financial-technology business in 2025, and became an AI lab in June 2026. Vinci names the AI research, models, technologies, and products developed by SimpleDirect.

Public model releases should be described according to their individual model cards. Model size, licence, evaluation coverage, and limitations differ by release.

Vero is Vinci’s character and human-facing explanatory layer. Vero does not certify or evaluate results.

Leadership

George Pu
Founder & CEO
Vikas Grover
COO & CFO
Ayush Naik
Technical Lead

Full leadership information

Media assets

For accurate editorial identification of SimpleDirect and Vinci. Keep marks unmodified and follow the brand rules. Model-family artwork is illustration, not a product screenshot or evaluation result. The Code image is an existing product capture; it is not a claim about the latest release.