Gemma 4 31B is Google DeepMind's most capable open-weight model, a 30.7-billion-parameter dense multimodal model released under the Apache 2.0 license. It processes text and image inputs with a 256K-token context window, supports configurable thinking/reasoning modes, native function calling, structured JSON output, and over 140 languages. Ranking among the top three open models globally on the Arena AI leaderboard, it matches or exceeds much larger models like Llama 4 and Qwen 3.5 on math, coding, and agent tool use, and can run quantized on consumer GPUs with 24GB of VRAM.