Gemma 4 12B IT vs Nemotron 3 Nano 4B

Gemma 4 12B IT

A mid-size cousin for agent work with tools, optional thinking and image and audio input, under Apache 2.0. Google's 4-bit build needs under 9 GB with working space, so it runs on a 12 GB GPU or any recent Mac in all the common runtimes. The main risks are no published safety results and no documented JSON output.

Nemotron 3 Nano 4B

A tiny reasoning cousin with trained tool calls, a thinking switch and 262,144 tokens of context, running in about 4 GB from NVIDIA's own GGUF on almost any machine. The main risks are serving commands that run repository code, no published safety results, a tool format left to the template, and files untouched and questions unanswered since March.

Gemma 4 12B IT

1. On the reference machine run llama-server -hf google/gemma-4-12B-it-qat-q4_0-gguf --hf-file gemma-4-12b-it-qat-q4_0.gguf --jinja -c 131072, and add --mmproj with mmproj-gemma-4-12b-it-qat-q4_0.gguf for images and audio
2. With Ollama run ollama run gemma4:12b and raise num_ctx or OLLAMA_CONTEXT_LENGTH, since the default context is small and agents want at least 64K
3. In vLLM or SGLang pass --tool-call-parser gemma4 --reasoning-parser gemma4 (vLLM also --enable-auto-tool-choice), or tool calls come back as text; Google's w4a16 build is the 4-bit route there
4. Thinking is off by default; pass enable_thinking=True or start the system prompt with <|think|>, and drop earlier turns' thinking from history except on tool-call turns
5. Put images before the text and audio after it, keep audio to 30 seconds and video to 60 frames, and validate tool names and arguments before running them
6. Transformers needs 5.10.0 or later for gemma4_unified, and mlx-lm has no gemma4_unified code, so on a Mac use mlx-vlm, LM Studio or Ollama

Nemotron 3 Nano 4B

1. Quickest start is ollama run nemotron-3-nano:4b, a 2.8 GB build
2. With llama.cpp use llama-server -hf nvidia/NVIDIA-Nemotron-3-Nano-4B-GGUF --jinja, NVIDIA's own Q4_K_M file
3. With vLLM, download nano_v3_reasoning_parser.py from the repository first and serve with --enable-auto-tool-choice --tool-call-parser qwen3_coder --reasoning-parser-plugin nano_v3_reasoning_parser.py --reasoning-parser nano_v3
4. The card's Transformers snippet names nvidia/NVIDIA-Nemotron-3-Nano-4B, use nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 instead
5. Thinking is on by default. Pass enable_thinking=False to the chat template for short tasks
6. Use temperature 1.0 and top_p 0.95 for reasoning and temperature 0.6 with top_p 0.95 for tool calling, as the card says
7. It's tuned for English only, and validate JSON yourself, since structured output isn't documented