One‑Click Deploy Gemma‑4‑31B with 256K Context, Matching Qwen 3.5 397B Performance

Gemma‑4 series overview

Google DeepMind open‑sourced the Gemma‑4 family of large language models under the Apache 2.0 license. The models share the same technology stack as Gemini 3 and rank in the top three on the Arena AI leaderboard while using far fewer parameters than competing models.

Model sizes and design rationale

The family includes multiple sizes—2 B, 4 B, 26 B and 31 B parameters—targeting different deployment scenarios. Smaller variants are optimized for lightweight, real‑time inference on edge devices; the 31 B variant is intended for high‑performance inference on powerful hardware.

31 B variant capabilities

The 31 B model achieves performance comparable to Qwen 3.5 397 B. It supports multimodal image‑text input, a context window of up to 256 K tokens, and native features for inference, function calling and system prompts. The model covers more than 140 languages, making it suitable for high‑quality question answering, code assistance and agent‑style services.

One‑click deployment on HyperAI

HyperAI provides a tutorial that clones the Gemma‑4‑31B‑it repository into a user‑owned container and runs it on a selected GPU image.

Open the HyperAI homepage, go to the “Tutorial” section, and select “Run this tutorial” for Gemma‑4‑31B‑it.

On the tutorial page, click the “Clone” button (top‑right) to copy the repository into the container.

Choose the “NVIDIA RTX PRO 6000” GPU image and a PyTorch environment, then click “Continue job execution”.

Wait until the job status changes to “Running”, then click “Open Workspace” to launch the Jupyter Workspace.

In the workspace, open the README file and click “Run”. After execution finishes, the displayed API endpoint can be used to query the model.

Running the notebook demonstrates successful model startup, multimodal input handling and API responses, confirming the functional correctness of the deployment.