> For the complete documentation index, see [llms.txt](https://support.vlex.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://support.vlex.com/vincent-by-vlex-1/vincent-studio/vincent-studio-advanced-techniques-and-best-practices/choose-the-ai-model-that-powers-a-studio-workflow.md). # Choose the AI Model That Powers a Studio Workflow Studio lets you select which large language model runs a workflow instead of leaving it to automatic selection, so you can control output consistency. #### Summary Studio lets you select which large language model runs a workflow instead of leaving it to automatic selection, so you can control output consistency before rolling the workflow out to your team. #### Why This is Important By default, Vincent decides which model to use for each task. That default is deliberate and it is good: it routes each step to a model suited to the work, and most builders never need to think about it. Some organizations do need to think about it. A firm standardizing a high-volume workflow across a hundred users cares less about peak quality on any single run than about getting the same shape of answer every time. A practice group lead who has validated a workflow's output wants confidence that the behavior they signed off on is the behavior their team gets. And an enterprise buyer evaluating Studio often needs to know *which* model is doing the work before they can approve the tool at all. Model selection answers all three. It moves a decision that used to sit with vLex into your hands, along with the responsibility for it. #### Before You Begin * You need access to **Studio** and permission to configure workflows. * Know what problem you are solving. Changing the model is a reasonable response to an observed, repeatable output issue. It is not a first troubleshooting step, and it is rarely the fix for a workflow whose prompts are underspecified. * Have a test document and a known-good expected output ready, so you can tell whether a change helped. #### How Automatic Selection Works Leaving the selector on **Default** means Vincent selects the best model for the work at hand. Vincent's architecture is model-agnostic by design: rather than being tied to one provider, it routes each stage of the question-answering process to a suitable model from a vetted lineup that includes OpenAI and Anthropic models. For most workflows this is the right setting, and it has a real advantage: as stronger models are added to the lineup, a workflow on **Default** benefits automatically, with no action from you. #### How to Change the Model 1. Open the model selector for the workflow. 2. Review the available options. Each model shows a short description of what it is suited for. 3. Select the model you want, or leave **Default** in place to keep automatic selection. 4. Save, then run the workflow against a test document and compare the output against your expected result. 5. To undo the change, reopen the selector and choose **Default** again.

The Studio model selector, showing Default alongside the available models.

#### Models Available in the Selector The selector offers **Default** plus a set of named models from OpenAI and Anthropic. Options at the time of writing: | Model | Suited for | | ------------------------ | ------------------------------------------------ | | **Default** | Vincent selects the best model for your workflow | | GPT-4o | Fast and versatile for everyday tasks | | GPT-4.1 | Large context for long documents | | GPT-5 | Strong reasoning for complex work | | GPT-5.1 | Balanced reasoning and speed | | GPT-5.2 | Deeper reasoning with extended effort | | GPT-5.4 | Frontier reasoning with large context | | GPT-5.5 | Frontier model that favours reasoning | | GPT-5.6 Sol, Terra, Luna | Newest frontier models with agentic tool use | | Claude 4.5 Sonnet | Balanced quality and speed | | Claude Opus 4.6 | Most capable, favours reasoning | | Claude Opus 4.8 | Highest quality for demanding tasks | {% hint style="info" %} **This list changes.** Models are added and retired as the market moves. Treat the selector itself as the source of truth for what is available to your organization, not this table. {% endhint %} #### What Changing the Model Does Not Change This is the part worth understanding before you start experimenting, because it narrows the set of problems model selection can actually solve. * **It does not change which legal authorities Vincent finds.** Vincent grounds its answers by first retrieving relevant documents from library's curated legal database, then instructing the model to answer only from those sources. Retrieval happens before the model does its work. Selecting a different model changes how the answer is reasoned through and written, not which authorities it is built on. * **It does not change your data protection posture.** All selectable models come from the same vetted provider lineup, covered by contractual zero data retention agreements: your prompts and documents are never used to train providers' models and are not retained by them beyond what is needed to process the request. * **It does not change your workflow's logic.** Your tasks, action types, prompts, and workflow assets are untouched. A workflow with a vague prompt produces vague output on every model in the list. * **It does not change jurisdiction coverage.** Research capability is bounded by the jurisdictions Vincent's research workflows support, regardless of which model generates the answer. #### Best Practices & Pro Tips * **Fix the prompt before you change the model.** When output quality disappoints, the cause is usually an underspecified prompt, a missing workflow asset, or a task doing too much at once. Those are cheap to fix and the fix compounds across every model. Switching models to solve a prompt problem hides the real defect and leaves you with a workflow that is now pinned to a specific model for no documented reason. * **Document why you moved off Default, in the workflow description.** Six months from now, someone will need to know whether the selection was a considered decision or an experiment nobody reverted. Without a note, the safe assumption is that it was an experiment, and the workflow gets rebuilt. * **Remember what pinning a model costs you.** A workflow on **Default** inherits improvements as the lineup evolves. A pinned workflow does not. If you pin one, put a reminder in your team's calendar to revisit the choice, and expect that any given named model will eventually be retired. ### Related Articles * [How to Publish and Share Your Vincent Studio Workflow](/vincent-by-vlex-1/vincent-studio/creating-and-managing-ai-workflows-in-vincent-studio/how-to-publish-and-share-your-vincent-studio-workflow.md) * [Best Practices for Planning an Effective Workflow](/vincent-by-vlex-1/vincent-studio/vincent-studio-advanced-techniques-and-best-practices/best-practices-for-planning-an-effective-workflow.md) ### Unlock the Power of Legal Automation Ready to see how Vincent Studio can transform your firm's efficiency? [**Contact our sales team for a personalized demo today.**](https://vlex.com/book-a-demo?utm_source=website\&utm_medium=organic\&utm_campaign=web_trial_homepage\&campaign_id=4270741) **What's Your Next Step?** | **New to Vincent Studio?** | **Already a Vincent Studio Customer?** | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- | | [See how Vincent can transform your practice. **Request a personalized demo**.](https://vlex.com/book-a-demo?utm_source=website\&utm_medium=organic\&utm_campaign=web_trial_homepage\&campaign_id=4270741) | [If you still have questions or need help with a specific feature, our team is here to assist. **Contact our Support team**.](https://vlex.com/contact-us) |