Google: Gemma 4 31B (free)

Gemma 4 31B Instruct is a Google DeepMind dense model with 30.7B parameters, built as an instruction-tuned system for multimodal understanding and text generation. Its long context window supports substantial documents and mixed inputs, while a configurable thinking mode lets developers trade additional reasoning work for deeper analysis. Native function calling and structured response support make it suitable for applications that combine document interpretation with tool-assisted workflows. The model is also described as supporting more than 140 languages, giving it broad multilingual reach for assistants, content workflows, and multilingual applications.

Available evaluations position this model as useful for coding, reasoning, and document understanding, with reported results of 85.7% on GPQA Diamond, 43.4 on the Artificial Analysis Coding Index, and 75.6% on IFBench in reasoning mode. Domain-focused testing also reports 61.4% on MortgageTax, 50.8% on Finance Agent, and 52.6% on CaseLaw. These results suggest a practical middle ground for long-context analysis, agentic coding, and professional document tasks rather than a single narrow specialization. In practice, it is a good fit for teams seeking open-weight, multilingual, multimodal assistance with adjustable reasoning and native tool integration.

Kilo Gatewaygoogle/gemma-4-31b-it:freegemma

Open provider page Provider docs

Quick Info

Provider: Kilo Gateway
Model key: google/gemma-4-31b-it:free
Release date: Apr 2, 2026
Last updated: Apr 2, 2026
Knowledge cutoff: 2025-01
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 32,768 tokens
Context window: 262,144 tokens

Latest news about Google: Gemma 4 31B (free)

Kilo Gateway

CoverageBenchmark

The Model Index spec page for Google: Gemma 4 31B (free) corroborates the same underlying model, describing Gemma 4 31B Instruct as Google DeepMind's 30.7B dense multimodal system with text and image input, text output, a roughly 262K token context window, a configurable thinking/reasoning mode, and native function calling. The page attributes creation to Google and marks the release date as April 1, 2026, one day earlier than the OpenRouter entry. Beyond the core specifications, the listing indicates the weights are available through an API for application integration and categorizes the release under a text-and-image input with text output modality set, with parameters left undisclosed on the card. The page conflicts with other registries by tagging the license as proprietary, so licensing and exact release date should be confirmed against Google's own documentation before publication.

Kilo Gateway

CoverageBenchmark

The OpenRouter listing for the exact variant google/gemma-4-31b-it:free describes Gemma 4 31B Instruct as Google DeepMind's 30.7B dense multimodal model handling text and image input with text output. It lists a 262K token context window, configurable thinking/reasoning mode, native function calling, and multilingual support across more than 140 languages, and states the model was released on April 2, 2026 under an Apache 2.0 license. The same page reports per-provider benchmark splits for the model, with GPQA Diamond scores ranging from 84.1% for the ModelRun entry up to 85.8% for SiliconFlow, and TAU-Bench Prime figures between 75.1% and 78.0% for the listed hosts. It also records tool-call and structured-output error rates and notes availability over the past three days, while provider routing and serving details remain gateway operations rather than model changes.