Gemma 4 12B IT vs ThinkingCap (Qwen3.8 27B)

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.

ThinkingCap (Qwen3.8 27B)

A Qwen3.8-27B with shorter reasoning, for agent work where thinking tokens cost time, and BottleCap's 4-bit GGUF runs in about 19 GB. The main limits are the licence, which needs a paid licence above 100 people or 1,000,000 USD revenue, a gated main repository, and no safety results or training-data description.

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

ThinkingCap (Qwen3.8 27B)

1. Check the licence first, an organisation with 100 or more people or 1,000,000 USD (2019) revenue needs a commercial licence from BottleCap
2. For files without the gate use llama-server -hf bottlecapai/ThinkingCap-Qwen3.8-27B-GGUF:Q4_K_M --jinja, about 17.4 GB
3. On vLLM use --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_xml, and on SGLang --tool-call-parser qwen3_coder
4. Set reasoning_effort (xhigh, medium or low) in chat_template_kwargs, any other value makes the template raise an error
5. Keep the base's sampling (temperature 1.0, top_p 0.95, top_k 20), the GGUF card says greedy decoding can loop
6. The GGUFs carry an MTP head, and llama.cpp builds without MTP support for this architecture may refuse to load them with a missing blk.64 tensor