gemma-4-31B-it is a 30.7-billion parameter dense multimodal model. It is built using the same research infrastructure as Gemini 3, offering frontier-level performance for reasoning, agentic workflows, and high-accuracy coding. It natively supports text and image inputs with a massive 256K context window.
The Gemma 4 31B-it is a 30.7-billion parameter dense multimodal model. It is built using the same research infrastructure as Gemini 3, offering frontier-level performance for reasoning, agentic workflows, and high-accuracy coding. It natively supports text and image inputs with a massive 256K context window.
To interact with the model via an OpenAI-compatible API (like NVIDIA NIM or vLLM):
| Parameter | Type | Required | Description |
|---|---|---|---|
model |
string |
Yes | Use "google/gemma-4-31b-it". |
messages |
array |
Yes | Array of objects with role and content. |
max_tokens |
integer |
No | Maximum tokens to generate. Default: 4096. |
temperature |
float |
No | Controls randomness. 0.3 for logic; 0.8 for creative. |
stream |
boolean |
No | Whether to stream tokens as they are generated. |
| Parameter | Type | Default | Description |
|---|---|---|---|
top_p |
float |
1.0 |
Nucleus sampling threshold. |
top_k |
integer |
50 |
Limits next token choices to top K probability tokens. |
enable_thinking |
boolean |
false |
Enables internal reasoning chains (if supported by provider). |
presence_penalty |
float |
0.0 |
Penalizes tokens based on whether they appeared. |
stop |
array |
null |
Custom sequences to stop generation. |
Run models at scale with our fully managed GPU infrastructure, delivering enterprise-grade uptime at the industry's best rates.
