DeepSeek R1 vs Qwen3: Which to Deploy Locally?

Two of the best Chinese open-source models—DeepSeek R1 and Qwen3—can both run locally. But which one should you install? That's a good question.

💡 What You Will Learn

Two of the best Chinese open-source models—DeepSeek R1 and Qwen3—can both run locally. But which one should you install? That's a good question.

|:--------|:--------|:--------| || Qwen3-7B (Q4_K_M) | 6GB | || DeepSeek R1-7B (Q4_K_M) | 6GB | || Qwen3 Coder-7B (Q4_K_M) | 6GB | || Qwen3-7B (Q4_K_M) | 6GB | || Qwen3-32B (Q4_K_M) | 20GB | || DeepSeek R1-7B | Qwen3-7B | Qwen3-Coder-7B | |:--------|:--------------|:---------|:--------------| || 62.3% | 68.3% | 65.1% | || 48.7% | 56.4% | 72.6% | || 82.5% | 76.8% | 71.2% | || 28.5 | 34.2 | 33.8 |

from openai import OpenAI

client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")

def smart_route(query):
    math_keys = ["", "", "", "", "", "", ""]
    code_keys = ["", "bug", "", "API", "debug", "", ""]

    if any(k in query for k in math_keys):
        model = "deepseek-r1:7b"
    elif any(k in query for k in code_keys):
        model = "qwen3-coder:7b"
    else:
        model = "qwen3:7b"

    resp = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": query}],
        temperature=0.7
    )
    return resp.choices[0].message.content, model

|:----|:----|:--------|:--------| | Qwen3-7B | Q4_K_M | 6GB | RTX 3060 12GB | | Qwen3-14B | Q4_K_M | 10GB | RTX 3090 24GB | | Qwen3-32B | Q4_K_M | 20GB | RTX 4090 24GB | | DeepSeek R1-7B | Q4_K_M | 6GB | RTX 3060 12GB |