- 🔒 Runs in a Trusted Execution Environment with hardware attestation
- 🆕 Google's Gemma 4 generation, dense 31B-parameter open model
- 🧠 Configurable thinking mode for step-by-step reasoning before answering
- 📏 This deployment exposes a 32K-token context window
- 👁️ Multimodal input: accepts both text and images
- 🌐 Pre-trained across many languages
- 🔧 Native function calling and system-prompt support
- 📚 Apache-2.0 licensed open weights
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 DeepMind and Google…
Explore 28 more models by Google →Gemma 4 31B Instruct is a member of Google's Gemma 4 open-weights family, here served inside a Trusted Execution Environment so that enclave identity and configuration can be independently verified via hardware attestation. The underlying model is a roughly 31B-parameter dense transformer with a vision encoder enabling text-and-image input. It is compact enough to run on capable single-GPU workstations.
Compared with its same-family predecessor, Gemma 3 27B, Gemma 4 introduces several concrete changes documented on Google's model card. It adds a built-in thinking mode for explicit step-by-step reasoning, native system-role support, and native function calling for agentic workflows. These features make it more directly usable for tool-driven and structured-output tasks than the earlier generation.
On safety, Google's model card describes improvements across content-safety categories relative to prior Gemma releases while aiming to keep unjustified refusals low. The model retains broad multilingual pre-training, and the 31B variant can process video supplied as frames.
Within Venice's catalog this instance carries a 32K context window and web-search plus end-to-end-encryption capabilities. Siblings include the non-enclave Google Gemma 4 31B Instruct and the Gemma 4 26B A4B Uncensored mixture-of-experts variant. Weights are released under Apache-2.0.
This About section is AI-generated from public sources via VeniceStats + Venice inference, with no human editing. It may contain inaccuracies.
| Seller | Rep↓ | Sales | Tokens | Routing | Input $/M | Cached $/M | Output $/M | Save |
|---|---|---|---|---|---|---|---|---|
| Leftermute 0x388b…5389 | 0.00 | — | — | gated | $0.0581 | $0.0581not advertised | $0.1742 | −58% |
| antseed-neon-puma-944e 0x6650…944e | 0.00 | — | — | gated | $0.0474 | $0.028 | $0.1466 | −66% |
"Best price" and the seller table are live Antseed catalog data (advertised $/1M tokens — or $ per generated image for unit-billed image models — not settled amounts). Reputation = buyer trust score (0-100, the Antseed SDK's own formula). "Routing" = the SDK's default buyer routing (what the VPR desktop app ships with): a trust ≥ 60 gate on the effective reputation, then cheapest-first among routable sellers; live failover state (per-peer cooldowns) is buyer-side runtime and not included. Model knowledge (TLDR, provider, About) via the VeniceStats enrichment layer. Advertised catalog, not the model used in any specific purchase. "Usage on Antseed" counts only settlements whose buyers share the per-model split on-chain (metadata v2/v3, opt-in), so every usage figure is a lower bound.