We are excited to announce the release of CPA-Qwen3-8B-v0, a specialized large language model designed specifically for the accounting and finance domain. Developed by AudCor, this 8-billion parameter model is fine-tuned to handle the rigorous logic required for Certified Public Accountant (CPA) tasks, regulatory compliance, and complex financial reasoning.
Why a Specialized Model?
General-purpose LLMs often struggle with the nuance required in professional accounting. They might hallucinate tax codes or miss the subtle distinctions in audit risk assessments.
CPA-Qwen3-8B was built to bridge that gap. By fine-tuning on high-quality financial datasets (including Josephgflowers/Finance-Instruct-500k), we have created a model that adopts the persona of a seasoned CPA—prioritizing professional skepticism, accuracy, and strict adherence to standards like GAAP and IFRS.
CPA Exam Insights
Understanding the standards against which this model is benchmarked requires looking at the rigorous CPA Exam scoring process.
Key Features
CPA Expert Persona
The model doesn't just answer questions; it adopts a professional tone suitable for drafting client emails, memos, or audit notes.
Exam-Grade Reasoning
Benchmarked against the logic required for the CPA Exam—known for its adaptive difficulty and "Gold Standard" passing score of 75—this model is capable of breaking down complex scenarios found in Audit (AUD), Financial Accounting and Reporting (FAR), and Regulation (REG).
Domain Specificity
It is optimized for specific tasks like audit risk assessment, tax planning strategies, and regulatory interpretation that often trip up generic models.
Technical Details
- Base Architecture: Built on the powerful Qwen3-8B architecture.
- License: Released under Apache 2.0, making it available for both academic and commercial use.
- Compatibility: Fully compatible with
transformersandunslothfor efficient inference and fine-tuning.
How to Use It
You can get started with the model immediately using the Hugging Face transformers library. Here is a quick example of how to set up the "CPA System Prompt" and ask a technical question about revenue recognition:
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "AudCor/cpa-qwen3-8b-v0"
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)
system_prompt = "You are an expert Certified Public Accountant (CPA). Your goal is to provide accurate, professional, and compliant financial advice..."
messages = [
{"role": "system", "content": system_prompt},
{"role": "user", "content": "Explain the revenue recognition principle under ASC 606."}
]
# Generate response code would follow here...
[!IMPORTANT] Disclaimer: While CPA-Qwen3-8B is a powerful research tool trained on high-quality financial data, it is not a licensed CPA. It should be used to assist professionals, not replace them. Always verify outputs for critical accounting, tax, or legal advice.
Get Involved
This is just version 0 (v0)! We invite the community to test the model, provide feedback, and contribute to the next generation of financial AI.
- Download the model: Hugging Face