Granite 4.2 3B vs Qwen3-4B-Instruct-2507

Granite 4.2 3B

A small reasoning cousin for tool calls, coding and multilingual chat that runs almost anywhere, from a 2.2 GB Ollama build to vLLM and SGLang, under plain Apache 2.0 with signed files and a dataset disclosure. Thinking turns off or shortens per request. It has no published safety results and no documented JSON mode.

Qwen3-4B-Instruct-2507

A small, fast cousin for instruction following, short tool-calling jobs and long documents on modest hardware, with no thinking step. It runs in about 3 GB at 4 bits on every common runtime, and two providers host it. The main risk is neglect, with no file changes in over a year, no answers from Alibaba and no safety results.

Granite 4.2 3B

1. Start it with ollama run granite4.2:3b; set options.num_ctx in the request, since Ollama's default context is short and the model takes up to 131,072
2. With vLLM 0.32.0 or later, serve ibm-granite/granite-4.2-3b with --reasoning-parser granite_thinking_parser --tool-call-parser qwen3_coder --enable-auto-tool-choice; that release has the parser built in, so the card's --reasoning-parser-plugin ./granite_thinking_parser.py isn't needed
3. With SGLang 0.5.18 or later, --reasoning-parser auto --tool-call-parser auto resolve to IBM's recommended parsers, and the reasoning comes back in reasoning_content
4. Use temperature 1.0 and top_p 0.95 for every mode, as IBM requires
5. Pass enable_thinking false for direct answers or low_effort true for short reasoning, through chat_template_kwargs on an OpenAI-compatible server
6. Earlier turns' thinking is stripped by default (truncate_history_thinking), so pass it false to keep reasoning in multi-turn history
7. For JSON, use the runtime's guided decoding, since IBM documents no JSON mode for 4.2
8. Verify the files with python -m model_signing verify sigstore and IBM's identity, per IBM's signature guide

Qwen3-4B-Instruct-2507

1. Quickest start is ollama run qwen3:4b-instruct, 2.5 GB. Plain qwen3:4b is the Thinking-2507 edition, not this one
2. On our reference machine use BF16 with vllm serve Qwen/Qwen3-4B-Instruct-2507 --max-model-len 262144 --enable-auto-tool-choice --tool-call-parser hermes
3. Use the hermes parser in vLLM or qwen25 in SGLang, the template emits JSON inside tool_call tags
4. Don't pass enable_thinking, this edition never thinks and the card says the flag isn't needed
5. Sampling per the card is temperature 0.7, top_p 0.8, top_k 20 and min_p 0, with presence_penalty up to 2 against repetition
6. Drop the context to 32,768 tokens if memory runs out, as the card suggests
7. Validate JSON yourself or use the runtime's guided decoding, since structured output isn't documented