Gemma 4 31B Instruct

• 🔒 Runs in a Trusted Execution Environment with hardware attestation for verification
• 🧠 Dense roughly 31B multimodal Gemma 4 model from Google, with thinking mode
• 👁️ Vision encoder accepts both text and image input
• 🆕 Adds native system-role support, which Gemma 3 lacked
• 🔧 Supports native function calling for agentic workflows
• 📏 This Venice deployment exposes a 32K context window
• 💬 Capabilities listed include reasoning and web search
• 📚 Apache-2.0 licensed, with millions of Hugging Face downloads

Google

Google is an American multinational technology corporation and one of the world's most valuable brands. A subsidiary of parent company Alphabet Inc., Google operates across search, cloud computing, consumer electronics, and artificial intelligence. Its…

Gemma 4 31B Instruct is a dense model from Google's Gemma 4 open-weights family, here packaged to run inside a Trusted Execution Environment (TEE) so that enclave identity and configuration can be independently verified through hardware attestation. The base model is a roughly 30.7B-parameter dense transformer paired with a vision encoder, accepting both text and image input; Google's model card describes it as multimodal, able to process video as frames.

Relative to its same-family predecessor Gemma 3 27B, Gemma 4 introduces several documented changes. It adds a configurable thinking mode for step-by-step reasoning, native function calling, and native system-role support, which Gemma 3 lacked. Google's model card also states that Gemma 4 models significantly improve over Gemma 3 and 3n on content-safety evaluations while keeping unjustified refusals low.

This Venice deployment lists a 32K context window and exposes reasoning and web-search capabilities. For other Google models in this catalog, see Google Gemma 4 31B Instruct and Gemini 3.1 Pro Preview.

The model is released under Apache-2.0 and reports strong adoption on Hugging Face, with millions of downloads, making it freely deployable for operators who want open weights combined with verifiable confidential execution.