Gemma 4 31B IT vs Olmo 3 7B Think

Gemma 4 31B IT

A dense multimodal cousin with documented tool calling, optional thinking and a 256K context, for general agent work and image reading on one 24 GB GPU through Google's 4-bit build. It runs in every major runtime under plain Apache 2.0. The main risk is that Google publishes no safety results and has gone quiet on the repository since July.

Olmo 3 7B Think

A fully open 7B reasoning cousin for maths, code and step-by-step answers, with every dataset and checkpoint published and its safety results rerunnable. Ollama's 4-bit build needs about 5.4 GB. It always thinks, wasn't trained for tool calls, has 65K of context, and Ai2 hasn't answered its discussions.

Gemma 4 31B IT

1. On the reference machine run llama-server -hf google/gemma-4-31B-it-qat-q4_0-gguf --hf-file gemma-4-31B_q4_0-it.gguf --jinja -c 32768, adding gemma-4-31B-it-mmproj.gguf for images
2. Or ollama run gemma4:31b for the 20 GB Q4_K_M build
3. In vLLM serve with --enable-auto-tool-choice --tool-call-parser gemma4 --reasoning-parser gemma4, as vLLM's recipe does
4. Thinking is off by default, so pass enable_thinking: true in chat_template_kwargs to turn it on, and expect the reasoning in the thought channel block
5. Drop earlier thoughts from the history except on tool-call turns, as the card asks
6. Use the card's sampling, temperature 1.0, top_p 0.95, top_k 64
7. Validate function names and arguments before running a call, and check JSON output yourself or use the runtime's guided decoding
8. The w4a16 build relies on Marlin INT4 kernels that vLLM marks unsupported on AMD, so use the GGUF on the reference machine

Olmo 3 7B Think

1. Quickest start is ollama run olmo-3:7b-think, about 4.5 GB
2. With vLLM, vllm serve allenai/Olmo-3-7B-Think --max-model-len 65536; vLLM runs it through its Transformers backend
3. Sample at temperature 0.6 and top_p 0.95 with up to 32,768 output tokens, as the card recommends, since thinking traces are long
4. Every reply starts with a think block; strip it yourself or use a reasoning parser that reads think tags
5. Pass your own system prompt; the template's default says the cut-off is November 2024 and that no functions are available
6. For tool calls use Olmo 3 7B Instruct instead, which Ai2 trained for function calling
7. A discussion reports endless reasoning loops when served with vLLM, unanswered, so cap max_tokens
8. Check the use against Ai2's Responsible Use Guidelines, which bar automated social-media posting and undisclosed automated content outside research