Google
Gemma 4 31B IT
Google's 31B-parameter open vision-language model with a 256K-token context, ready to serve on Runpod Serverless with vLLM or on a Pod.
Gemma 4 31B IT
Specifications
Provider
Google
Modality
Vision + language
Task
Text generation
Parameters
31B
Active parameters
Context
256K tokens
Precision
BF16
Weights
62
License
apache-2.0
Released
2026-03-10
Endpoint ID
Weights on Hugging Face
google/gemma-4-31B-it
Frequently asked questions
How do I deploy Gemma 4 31B IT on Runpod?
Deploy the vLLM worker from the Runpod Hub as a Serverless endpoint and set MODEL_NAME to google/gemma-4-31B-it. The endpoint is OpenAI-compatible and scales to zero when idle. You can also run vLLM on a Pod.
What GPU do I need to run Gemma 4 31B IT?
Start with H100 (80 GB VRAM). The weights take about 62 GB; longer contexts and more concurrent requests need extra memory for the KV cache.
How much does Gemma 4 31B IT cost on Runpod?
Serverless endpoints and Pods bill per second for the GPU you choose. A Serverless endpoint costs nothing while scaled to zero.
google-gemma-4-31b-it
google/gemma-4-31B-it
Vision + language
2026-03-10 17:00
H100