BCCard/Qwen3-Coder-30B-A3B-Instruct-FP8-Dynamic

Comparison between Qwen3-Coder-30B and Qwen3-235B Models

Performance Comparison

Benchmark	Qwen3-Coder-30B-A3B-Instruct-FP8-Dynamic	Qwen3-235B-A22B	Qwen3-235B-A22B-FP8-dynamic
MMLU	78.07%	84.77%	84.61%
ARC Challenge	65.36%	71.84%	70.90%
GSM-8K (strict-match)	88.93%	74.22%	74.98%
Hellaswag	50.25%	76.56%	76.10%
Winogrande	62.75%	73.95%	75.06%
TruthfulQA MC2	59.00%	61.18%	60.93%

Key Observations

1. GSM-8K Performance: Qwen3-Coder-30B significantly outperforms the larger Qwen3-235B models on GSM-8K (mathematical reasoning), with a score of 88.93% compared to ~74% for the 235B models.

2. Hellaswag Performance: The 30B model shows a significant drop in performance on Hellaswag (commonsense reasoning), scoring only 50.25% compared to ~76% for the 235B models.

3. TruthfulQA Performance: All models perform similarly on TruthfulQA, with the 30B model slightly behind at 59.00% compared to ~61% for the 235B models.

4. Winogrande Performance: The 30B model scores 62.75% on Winogrande compared to ~74-75% for the 235B models.

Summary

The Qwen3-Coder-30B model shows specialized strength in mathematical reasoning tasks (GSM-8K) where it significantly outperforms the larger models. However, it lags behind the 235B models in general knowledge (MMLU), commonsense reasoning (Hellaswag), and pronoun disambiguation (Winogrande). This suggests the 30B model may have been optimized specifically for coding and mathematical tasks at the expense of some general capabilities.

---

Qwen3-Coder-30B-A3B-Instruct

Highlights

Qwen3-Coder is available in multiple sizes. Today, we're excited to introduce Qwen3-Coder-30B-A3B-Instruct. This streamlined model maintains impressive performance and efficiency, featuring the following key enhancements:

• Significant Performance among open models on Agentic Coding, Agentic Browser-Use, and other foundational coding tasks.
• Long-context Capabilities with native support for 256K tokens, extendable up to 1M tokens using Yarn, optimized for repository-scale understanding.
• Agentic Coding supporting for most platform such as Qwen Code, CLINE, featuring a specially designed function call format.

Model Overview

Qwen3-Coder-30B-A3B-Instruct has the following features:

• Type: Causal Language Models
• Training Stage: Pretraining & Post-training
• Number of Parameters: 30.5B in total and 3.3B activated
• Number of Layers: 48
• Number of Attention Heads (GQA): 32 for Q and 4 for KV
• Number of Experts: 128
• Number of Activated Experts: 8
• Context Length: 262,144 natively.

NOTE: This model supports only non-thinking mode and does not generate <think></think> blocks in its output. Meanwhile, specifying enable_thinking=False is no longer required.

For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our blog, GitHub, and Documentation.

Quickstart

We advise you to use the latest version of transformers.

With transformers<4.51.0, you will encounter the following error:

KeyError: 'qwen3_moe'

The following contains a code snippet illustrating how to use the model generate content based on given inputs.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen/Qwen3-Coder-30B-A3B-Instruct"

# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

# prepare the model input
prompt = "Write a quick sort algorithm."
messages = [
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

# conduct text completion
generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=65536
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist() 

content = tokenizer.decode(output_ids, skip_special_tokens=True)

print("content:", content)

Note: If you encounter out-of-memory (OOM) issues, consider reducing the context length to a shorter value, such as 32,768.

For local use, applications such as Ollama, LMStudio, MLX-LM, llama.cpp, and KTransformers have also supported Qwen3.

Agentic Coding

Qwen3-Coder excels in tool calling capabilities.

You can simply define or use any tools as following example.

# Your tool implementation
def square_the_number(num: float) -> dict:
    return num ** 2

# Define Tools
tools=[
    {
        "type":"function",
        "function":{
            "name": "square_the_number",
            "description": "output the square of the number.",
            "parameters": {
                "type": "object",
                "required": ["input_num"],
                "properties": {
                    'input_num': {
                        'type': 'number', 
                        'description': 'input_num is a number that will be squared'
                        }
                },
            }
        }
    }
]

import OpenAI
# Define LLM
client = OpenAI(
    # Use a custom endpoint compatible with OpenAI API
    base_url='http://localhost:8000/v1',  # api_base
    api_key="EMPTY"
)
 
messages = [{'role': 'user', 'content': 'square the number 1024'}]

completion = client.chat.completions.create(
    messages=messages,
    model="Qwen3-Coder-30B-A3B-Instruct",
    max_tokens=65536,
    tools=tools,
)

print(completion.choice[0])

Best Practices

To achieve optimal performance, we recommend the following settings:

1. Sampling Parameters:

   • We suggest using temperature=0.7, top_p=0.8, top_k=20, repetition_penalty=1.05.

2. Adequate Output Length: We recommend using an output length of 65,536 tokens for most queries, which is adequate for instruct models.

Citation

If you find our work helpful, feel free to give us a cite.

@misc{qwen3technicalreport,
      title={Qwen3 Technical Report}, 
      author={Qwen Team},
      year={2025},
      eprint={2505.09388},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.09388}, 
}