MiniCPM5-2B vs Qwen3-Coder-30B-A3B-Instruct

MiniCPM5-2B

A tiny dense cousin for local tool calling and long context, with published training data and an active publisher. OpenBMB's own 4-bit builds need about 2 GB, so it runs on a laptop or phone in vLLM, SGLang, llama.cpp, Ollama or Transformers. The main risks are always-on thinking, no safety results and use conditions beyond Apache 2.0.

Qwen3-Coder-30B-A3B-Instruct

A fast coding cousin, with 3.3B of 30.5B parameters active per token, for agentic coding tools, repository reading and fill-in-the-middle without a thinking step. It needs about 21 to 23 GB at 4 bits, and several providers host it. The main risk is that it runs commands for agents with no safety results or guidance, and Alibaba doesn't answer users.

MiniCPM5-2B

1. Smallest start is ollama run openbmb/minicpm5-2b (1.6 GB, q4_k_m)
2. On vLLM 0.29.0 or later run vllm serve openbmb/MiniCPM5-2B --max-model-len 131072 --enable-auto-tool-choice --tool-call-parser minicpm5, or tool calls come back as XML text
3. On SGLang 0.5.16 or later add --tool-call-parser minicpm5 --reasoning-parser qwen3; SGLang is OpenBMB's recommended backend for tool calls
4. Tool calls are XML, <function name=...><param name=...>, not Hermes JSON
5. Sample with temperature 1.0, top_p 0.95 and min_p 0.0; llama.cpp's default min_p 0.05 can loop, and repetition_penalty 1.05 helps if it does
6. Expect a <think> block before every answer and leave room in max_tokens; enable_thinking false is accepted by the template but OpenBMB doesn't support it for the 2B
7. Validate JSON yourself or use the runtime's guided decoding, since structured output isn't documented
8. Use the main repository's template; the GGUF, MLX and GPTQ copies predate the 29 September fix

Qwen3-Coder-30B-A3B-Instruct

1. Quickest start is ollama run qwen3-coder:30b, 19 GB
2. On our reference machine use BF16 with vllm serve Qwen/Qwen3-Coder-30B-A3B-Instruct --max-model-len 262144 --enable-auto-tool-choice --tool-call-parser qwen3_coder, 61 GB of weights fits the GPU share
3. Use the qwen3_coder parser in vLLM or SGLang, not hermes, since calls come back as function and parameter blocks rather than JSON
4. Pass tool-call arguments in the history as objects, the template iterates them as a mapping
5. Sampling per the card is temperature 0.7, top_p 0.8, top_k 20 and repetition_penalty 1.05, with up to 65,536 output tokens
6. For code completion use the fill-in-the-middle tokens <|fim_prefix|>, <|fim_suffix|> and <|fim_middle|> from Qwen's README
7. Sandbox the commands and files it touches, since nothing from Alibaba covers prompt injection