It’s that time of the year again and tax filing is indeed a taxing process! Are you struggling with filing your ITR? Or are you still relying on your personal CA to handle it all? How about having a personalized AI tax assistant at your fingertips? With powerful AI models like OpenAI’s o3-mini on ChatGPT, and DeepSeek’s R1, tax filing is now easier and faster than ever. But which AI model does a better job at handling deductions, understanding new tax laws, and optimizing your savings? In this blog, we put o3-mini and DeepSeek-R1 to the test with real-world tax scenarios – involving salary slips, rent receipts, and tricky deductions – to see who’s the better AI accountant.
New Tax Regime 2025-26
In the Union Budget 2025, Finance Minister Nirmala Sitharaman has introduced significant changes to the income tax slab under the new regime. These new slabs will be effective from April 1,2025 for the financial year 2025-26. This budget’s tax structure is designed to majorly benefit employed taxpayers, especially in the middle income category. Here are the major take aways from Budget 2025.
- Standard Deduction: The standard deduction for salaried individuals has been raised from ₹50,000 to ₹75,000, providing additional tax relief.
- Increased Tax Exemption Limit: Individuals earning up to ₹12 lakh annually are now exempted from paying income tax. For salaried individuals, after accounting for a standard deduction of ₹75,000, this exemption effectively applies to incomes up to ₹12.75 lakh.
- Revised Tax Income Slabs: Here’s the table for revised tax slabs, according to the new regime.
| Annual Income (₹) | Tax Rate |
| 0 – 4 Lakh | Nil |
| 4 – 8 Lakh | 5% |
| 8 – 12 Lakh | 10% |
| 12 – 16 Lakh | 15% |
| 16 – 20 Lakh | 20% |
| 20 – 24 Lakh | 25% |
| Above 24 Lakh | 30% |
Are you feeling overwhelmed by all of these changes? Are you worried about remembering all the details and planning your taxes correctly? No worries! Why stress when you can have an AI tax assistant do it for you?
DeepSeek-R1 vs OpenAI o3-mini for Tax Filing
You’ve probably heard about the Chinese AI model DeepSeek-R1 and its biggest American competitor, OpenAI’s o3-mini. Although both these AI models are known to be exceptional at logical reasoning, mathematics, and more, can they actually help with tax calculations? Let’s put them to the test!
We’ll use DeepSeek-R1 and o3-mini as AI tax assistants and explore two different scenarios to see how well they handle financial planning. But before that, let’s see how to access these models.
How to Access o3-mini?
Step 1: Head to https://chatgpt.com/ and select o3 mini from the top left drop down of model options.
Step 2: Give the prompt.
Also Read: How to Run OpenAI’s o3-mini on Google Colab?
How to Access DeepSeek-R1?
Step 1: Go to https://www.deepseek.com/ and select ‘Start now’. Then sign in using your Gmail account.
Step 2: Select DeepSeek-R1 and then provide the prompt.
Now let’s get to the tax filing.
Scenario 1: Which Model Has Better Understanding of the New Regime FY 2025-26?
Prompt: “Here’s a link to the new tax regime proposed in Budget 2025: https://economictimes.indiatimes.com/wealth/income-tax-slabs?from=mdr.
Using only the information in that article, what is the standard deduction for the financial year 2025–26”
Response by o3-mini (high)
Response by DeepSeek-R1
Comparative Analysis
You can clearly see that OpenAI ChatGPT o3-mini gave the wrong answer, while DeepSeek-R1 got it right. Under the new tax regime, the standard deduction for a salaried employee is ₹75,000. This shows that DeepSeek-R1 could collect the relevant information from the article better than ChatGPT o3-mini.
Also Read: Can o3-mini Replace DeepSeek-R1 for Logical Reasoning?
Scenario 2: Which Model Can Handle the New Tax Regime Better?
Let’s Take a Test Case: Priya Mehta
- Age: 29
- Salary: ₹13,00,000 per annum
- Deductions & Expenses:
- Home Loan Interest: ₹1,20,000 per year
- Medical Bills: ₹30,000 (eligible under 80D)
- Equity Investments: ₹1,50,000 (eligible under 80C)
Priya is a salaried employee with multiple expenses. She has shown her investments under 80C (Equity Investments). 80D (medical expenses), and 80E (education loan interest) while ensuring compliance with new tax laws for 2025.
Let’s manually calculate Priya’s tax under the new tax regime.
In this regime, deductions under 80C and 80D are not allowed. So, if we calculate the tax for her annual salary of ₹13 lakh, the total tax payable will be as follows:
Gross income is = ₹13,00,000 – ₹75,000 = ₹12,25,000
| Annual Income (₹) | Tax Rate | Payable Tax |
| 0 – 4 Lakh | Nil | 0 |
| 4 – 8 Lakh | 5% | ₹20,000 |
| 8 – 12 Lakh | 10% | ₹40,000 |
| 12 – 12.25 Lakh | 15% | ₹3,750 |
| Total Tax payable | ₹ 63,750 |
Add Health Cess
4% of ₹ 63,750= ₹2,550
| Total Tax = ₹ 63,750 + ₹2,550 = ₹ 66,300 |
Now that we’ve manually calculated Priya’s tax, let’s see if these AI models can do the same accurately for this financial year and compare their performance.
Prompt: “Here’s a link to the new tax regime proposed in Budget 2025: https://economictimes.indiatimes.com/wealth/income-tax-slabs?from=mdr.
Using only the information in that article, compute Priya Mehta’s tax liability under the new tax regime proposed in the 2025 Budget. Below is a detailed breakdown of her salary and expenses.
Name: Priya Mehta
1. Age: 29
2. Salary: ₹13,00,000 per annum
3. Deductions & Expenses:
Home Loan Interest: ₹1,20,000 per year
Medical Bills: ₹30,000 (eligible under 80D)
Equity Investments: ₹1,50,000 (eligible under 80C)
Priya is a salaried employee with multiple expenses. She has shown her investments under 80C (Equity Investments). 80D (medical expenses), and 80E (education loan interest) while ensuring compliance with new tax laws for 2025.”
Response by o3-mini (high)
Response by DeepSeek-R1
Comparative Analysis
ChatGPT, powered by o3-mini, applied a standard deduction of ₹50,000 but didn’t allow deductions under 80C, 80D, or home loan interest. This brought Priya’s taxable income down to ₹12,50,000 (₹13,00,000 – ₹50,000). However, it used the old tax slabs, which led to an incorrect final tax liability of ₹1,30,000.
DeepSeek-R1 took more time to analyze and understand the tax regime but still made an error in the calculation. It missed deducting the standard ₹75,000 from Priya’s total income. Then, instead of applying the new tax regime for 2025, it used the old 2023-24 rules, which led to an incorrect tax liability of ₹62,400.
Also Read: This o3-mini Agent Can Predict Gold Prices!
Conclusion
Both ChatGPT with o3-mini, and DeepSeek-R1 show promise in AI tax filing. However, each struggled with certain details such as understanding the standard deductions and new tax slabs. While these AI-powered tax solutions can streamline filing and reduce manual effort, they’re not yet perfect substitutes for expert oversight. As AI continues to evolve, these models will most likely become more accurate in accounting and tax filling.
Frequently Asked Questions
A. AI tax filing uses advanced language models like o3-mini on ChatGPT, and DeepSeek-R1, to help individuals and businesses prepare, calculate, and submit tax returns with minimal manual effort.
A. AI can quickly process large sets of financial data, offer real-time updates on tax laws (like the 2025–26 new tax regime), and automate repetitive tasks. However, expert review may still be advisable.
A. Key updates include a higher tax exemption limit (up to ₹12.75 lakh with standard deduction), revised slabs, and an increased standard deduction of ₹75,000 for salaried individuals.
A. In theory, yes, but the blog’s scenarios revealed that both AI models sometimes missed or misapplied these deductions under the new regime rules.
A. They’re useful for guiding basic calculations and saving time. However, they can still misunderstand nuanced or newly introduced laws. Hence, human oversight is recommended.
A. AI can handle basic calculations and tax queries, but it can’t fully replace a tax accountant’s nuanced judgment and personalized expertise – especially in complex or high-value situations.
A. According to this article, neither model got it entirely right. o3-mini used outdated deductions, while DeepSeek-R1 applied old tax slabs, leading to inaccuracies.
As an Instructional Designer at Analytics Vidhya, Diksha has experience creating dynamic educational content on the latest technologies and trends in data science. With a knack for crafting engaging, cutting-edge content, Diksha empowers learners to navigate and excel in the evolving tech landscape, ensuring educational excellence in this rapidly advancing field.
