How to upgrade your LLM provider from Phi 4 to Gemma 4
How to upgrade your LLM provider from Phi 4 to Gemma 4
Introduction
We're upgrading the LLM model behind M42 GenAI from Phi-4 to Gemma 4 — a free upgrade available to all customers today. This is a straightforward change on our side: no additional cost, no new contract, no extra licensing. Just an improved model powering the same AI features you already use.
You'll notice that the provider setting in the new configuration is labeled "OpenAI." This is just a naming convention in the interface — it does not mean your data is sent to OpenAI. In reality, both the current provider (M42 GenAI (legacy), running Phi-4) and the upgraded provider (labeled OpenAI, running Gemma 4) point to M42 GenAI, the model hosted and operated by Matrix42. The "OpenAI" label simply reflects the API notation/protocol used to connect to the model — the endpoint URLs below confirm this, both pointing to Matrix42-hosted infrastructure.
This guide walks you through checking your current provider setting and upgrading to Gemma 4.
Data processing and residency
This upgrade does not change how or where your data is processed. Gemma 4 runs on the same M42 GenAI infrastructure, hosted and operated by Matrix42, as the current Phi-4 model — there is no change to hosting location, data residency, or the parties processing your data. Because nothing changes on this front, no action is required on your side regarding data processing agreements, internal compliance reviews, or policy updates related to this upgrade.
How to check which provider you're currently using
Navigate to the M42 Intelligence tab in the Service Management Platform's Admin UI → Provider settings.
The following settings indicate that you are using Phi-4:
- Select generative AI provider:
M42 GenAI (legacy) - AI provider url:
http://10.210.228.1:8081/api/chat/completion
In some cases, the provider settings can also point to an URL that begins with “Puna”.
How to upgrade to the latest LLM model
Change the settings accordingly:
- Select generative AI provider:
OpenAI - AI provider API password: [API key provided by Matrix42 – contact your PSO consultant, partner, or Service Desk]
- AI provider API URL: [API URL provided by Matrix42 – contact your PSO consultant, partner, or Service Desk]
- Model:
gemma4
After changing the settings, ensure that the connection and use cases work:
- Check connectivity by clicking “Test connection”. You should see a “Connection successful” banner if the connection works.
- Test your Actions and/or Writing assistance use cases
ℹ️ Prompt changes? Good news — barely any.
Gemma 4 is a significant step up from Phi-4, and it shows in how little you need to adjust:
- Your existing prompts will work well as-is. Gemma 4 handles them with better accuracy and more consistent results, no rewrite needed.
- You can actually simplify your prompts over time. Gemma 4 is smarter and more autonomous, so it needs less hand-holding and fewer explicit instructions to get the right output — a good practice to start adopting.
- Broader language support, so more of your users can get quality results in their own language.
In short: same prompts, better results — with room to streamline further whenever you're ready.
Table of Contents