Skip to main content
Vinci MLE 1.0 · Research release

Build your own ML engineering capability.

AI that helps your team build AI. Vinci MLE by SimpleDirect® is an open-weight research family for government and enterprise ML teams exploring experiment diagnosis and bounded engineering changes on infrastructure they control.

MLE 1.0 is a research release. A valid held-out ML-engineering task score is not yet available.

Vero in a warm sage and cream setting illustrates the model family’s work.
A conceptual Vero scene, not a model run or evaluation result.

Vinci MLE

Keep the experiment, evidence and decisions in your hands.

Developing AI takes more than access to a model. Teams need repeatable experiments, well-understood data and people who can judge the result. Vinci MLE is intended to support that work while your organization builds its own expertise.

Available today

A research model you can inspect

Published 8B, 30B and 123B weights and model cards describe the training, foundations, licences and measured general-capability results. Tools, permissions and the execution workspace must be provided and validated separately.

Scoped engineering

Engineering around your ML workflow

Discuss a scoped project around your experiment environment: deployment, representative task sets, acceptance checks, integration or adaptation. Determine whether the model is suitable before relying on it or expanding access.

Future direction

An evidence-led engineering assistant

Our direction is to connect specialist models to controlled experiment workflows, with traceable actions and clear stop conditions. Held-out ML-task evaluation and a qualified execution environment are prerequisites for stronger capability claims, not completed 1.0 results.

Released weights are the starting point. A private deployment, integration or adapted model is a separately scoped engineering project. Future capabilities are goals, with availability and acceptance criteria still to be established.

Help with the work of building AI

Vinci MLE is trained to read files and experiment results, suggest a limited fix when evidence supports it, preserve a working setup when it does not, and identify missing evidence instead of guessing.

ML engineers and AI researchers can evaluate it on experiment diagnosis, training recovery, memory and batch settings, scoring checks, model packaging and configuration consistency. These describe intended workflows and training coverage, not demonstrated performance on new tasks.

Illustrative evaluation workflow. These steps describe a trial to build, not a live product demonstration.

  1. Define one experiment problem

    Use a small, non-sensitive workspace with an experiment configuration, observed results and a clear question, such as why a run exhausted memory.

  2. Ask for a bounded diagnosis

    Evaluate whether the model uses the supplied evidence, identifies what is missing and proposes a limited change, a justified no-change decision or a stop.

  3. Check the result independently

    Have an engineer review the proposal, then rerun agreed checks in an isolated environment. Compare with the original configuration and a baseline; a completed run alone is not correctness.

SimpleDirect® specializes existing IBM Granite and Mistral Devstral foundations through post-training. These releases are not foundation models pretrained from scratch in Canada.

Choose a model for your environment

8B has the smaller weight download; 30B has a larger footprint. 123B uses Devstral 2 and a different licence. Use each detail page for weight-file sizes and deployment notes. General coding results do not rank these models on ML-engineering tasks.

Published release 1.0

8B

A smaller starting point

The smallest weight footprint in this family. Consider it for an initial local workflow, then measure quality, latency and memory on your own task.

Explore Vinci MLE 8B 1.0

Published release 1.0

30B

A larger Granite option

Compare this IBM Granite adaptation with 8B on the same cases. Plan for a larger weight footprint; a larger model does not guarantee a better result.

Explore Vinci MLE 30B 1.0

Published release 1.0

123B

A different foundation

A Mistral Devstral adaptation with different infrastructure and licence considerations. Check deployment fit and the company-revenue condition before choosing it.

Explore Vinci MLE 123B 1.0

B indicates the approximate number of parameters in billions, not a quality score. Memory needs also depend on file format, context length, simultaneous requests and runtime; weight-file size is not a complete memory requirement.

Scroll the table horizontally for licences and downloads.

Licences and available formats are artifact-specific. Follow each model card for the detailed licence, loading examples and file checks.

The 8B and 30B releases use Apache-2.0. The 123B releases use Mistral’s Modified MIT licence, which includes a company-revenue condition. Review the linked terms before choosing a model for your organization.

Separate training intent from evaluation evidence

Valid held-out ML-engineering and external MLE-bench evaluations were not completed for 1.0. General-capability comparisons with each model’s own parent document measured gains and tradeoffs, but do not establish performance on new ML-engineering tasks.

Use each artifact’s own evaluation record. Results for a parent model, sibling size or full-precision checkpoint do not establish the performance of a different release or quantized format.

Plan a self-hosted deployment

Start with a bounded experiment, a baseline and explicit acceptance criteria. Download full weights or a supported GGUF format from the model card, then evaluate suggestions against the original experiment evidence before permitting file edits or commands.

Allow memory beyond the weight files, choose a supported runtime, and record the exact revision and format. Tool use requires your own application, permissions and controlled workspace. Review and test suggested changes before applying them; these pages do not offer hosted inference.

Decide where information can go
Map model hosting, storage, logs, backups and every connected tool. Confirm data classification, location, access and retention requirements before introducing sensitive information.
Keep permission to act explicit
Begin in a restricted test workspace. Separate suggestions from execution, limit tool access, require approval for consequential changes and define a stop or rollback path.
Make results reproducible
Pin model and dataset revisions, runtime, format and prompts. Compare with a baseline on representative cases, including correct inputs, failures and cases where the model should stop.
Build knowledge your team can keep
Agree on documentation, evaluation assets, operating responsibilities, licensing and handover. Ownership and adaptation rights depend on the underlying licences and the project agreement.

For government and regulated work, your organization must assess the deployment and approve it for the information involved. Download availability does not establish security accreditation or suitability for a particular data classification.

From published weights to your own capability

Use the published weights independently under their licences. If you need engineering help, shape a project around your requirements.

  1. Define the useful outcome

    Bring a concrete workflow and deployment requirements. Request implementation or adaptation directly, or scope a separate evaluation if suitability is still uncertain.

  2. Agree the delivery

    Set deliverables, acceptance checks, data rights, infrastructure ownership and procurement requirements. Confirm feasibility and responsibilities before implementation.

  3. Plan handover and support

    Decide who owns documentation, updates, monitoring and incident handling. Any ongoing support or operations are separately scoped, with availability and responsibilities agreed for the project.

Start with one useful workflow

Start with one experiment your team understands.

Describe your ML-engineering task without sharing confidential material and ask for an implementation or adaptation proposal. A separate evaluation engagement is optional; deployment still needs agreed acceptance checks.

Opens your email app with an editable outline to [email protected]. Nothing is submitted by this page.

A useful first conversation

  • Which experiment or ML-engineering task should the project address?
  • What compute, tools and review process does your team use?
  • Which baseline, reproducibility checks and stop conditions would make the result meaningful?

Keep the first message non-confidential. Do not send source code, security findings, documents, personal data or credentials. Agree on an appropriate sharing process first.