AvenChat is a free web platform for exploring, comparing, and testing Google's Gemma 4 AI model family, offering interactive chat, hardware guides, and local setup tutorials.
What is AvenChat?
AvenChat is a browser-based hub for Google's Gemma 4 open-weight model family, providing a free chat interface to test prompts against Gemma 4 variants and a library of decision guides. It takes user text, image, or video prompts and returns model-generated outputs, backed by detailed specifications for all five official Gemma 4 sizes (E2B, E4B, 12B, 26B A4B, and 31B). The platform runs entirely in the web browser and is maintained by AvenChat (not Google), with guides for local deployment, hardware planning, and model comparison.
Key Features
- Free Web Chat — Test Gemma 4 models directly in your browser with persistent conversations (sign-in required to save).
- Model Comparison Guides — Side-by-side analysis of all five variants: 31B (dense, quality-first), 26B A4B (MoE for efficiency), 12B (balanced multimodal), E4B, and E2B (lightest), covering context length (128K–256K), multimodal support, and approximate Q4 memory (2.9 GB to 17.5 GB). Compare the models
- Hardware Requirement Guides — Clear memory specifications per model and quantization, helping you choose before downloading. Check hardware needs
- Local Setup Tutorials — Step-by-step instructions for running Gemma 4 in Ollama, LM Studio, llama.cpp, MLX, and Gemma.cpp. See setup options
- Multimodal Support — All models accept image and video input; E2B, E4B, and 12B also support native audio input.
- Apache 2.0 License — Commercially permissive, enabling self-hosting, customization, and product integration.
Who is it for?
- AI enthusiasts — Test Gemma 4's reasoning and multimodal capabilities for free before committing to local setup.
- Developers evaluating models — Compare Gemma 4 variants against each other and alternatives like Qwen using specific benchmarks and context windows.
- Teams planning deployment — Use hardware guides and local setup tutorials to select the right model for their infrastructure and budget.
- Researchers and analysts — Leverage long-context (up to 256K tokens) and multimodal input for document analysis, agent workflows, and video summarization.
What can you do with it?
- Test prompts — Use the free chat to pressure-test prompts, summarize documents, and compare outputs across Gemma 4 sizes. Try the free chat
- Choose a model — Read the family comparison to decide between dense, MoE, and lightweight variants based on your hardware and quality needs.
- Plan local hardware — Consult the hardware guide for exact RAM/VRAM requirements per quantization (e.g., Q4 12B ~6.7 GB, 31B ~17.5 GB).
- Set up locally — Follow the Ollama or LM Studio guides to run Gemma 4 on your own machine. Open Ollama guide
FAQ
What is Gemma 4?
Gemma 4 is Google's open-weight model family built for reasoning, multimodal input, and flexible deployment, available in five sizes from E2B (2.9 GB Q4) to 31B (17.5 GB Q4), all under Apache 2.0 license.
Is AvenChat free to use?
Yes. AvenChat provides a free browser-based way to try Gemma 4 without any payment or sign-up for basic chat (though sign-in is required to persist conversations).
Can I run Gemma 4 locally?
Yes. Gemma 4 is designed for local deployment via Ollama, LM Studio, llama.cpp, MLX, and Gemma.cpp, with setup guides available on AvenChat.
What hardware do I need for Gemma 4?
Hardware depends on model and quantization. Approximate Q4 memory: E2B ~2.9 GB, E4B ~4.7 GB, 12B ~6.7 GB, 26B A4B ~14.7 GB, 31B ~17.5 GB. See the hardware guide for details.
Does Gemma 4 support images and audio?
All models accept image and video input. E2B, E4B, and 12B additionally support native audio input. 31B and 26B A4B focus on text-plus-visual workloads.
