OpenAI released GPT-5.2 with native spreadsheet analysis

OpenAI released GPT-5.2 this week, upgrading its flagship model with native spreadsheet analysis and faster processing speeds. For UK SME owners, the update makes preparing messy financial data much easier, but you still can't trust it to reconcile your accounts without human oversight. The new model handles large CSV files seamlessly, yet it continues to struggle with the deterministic reliability strict bookkeeping requires.

OpenAI ships GPT-5.2 with native spreadsheet tools

OpenAI launched GPT-5.2 in December 2025, introducing a model that natively processes complex spreadsheets and large datasets Reuters (https://www.reuters.com/technology/openai-launches-gpt-52-ai-model-with-improved-capabilities-2025-12-11). The update integrates advanced data analysis directly into the core chat interface, allowing users to upload heavy Excel or CSV files and ask the system to clean, sort, and visualise the contents.

Unlike previous versions that relied on a separate code interpreter environment to run Python scripts for every calculation, GPT-5.2 attempts to read and calculate tabular data natively. This makes the model significantly faster at generating charts and summarising trends. However, developers testing the system note that while it excels at formatting and exploratory analysis, it still hallucinates specific cell values when handling large files OpenAI Developer Community (https://community.openai.com/t/accuracy-of-retrieved-data-large-csv/715494). The model understands the structure of a profit and loss statement, but it doesn't apply the rigid, rule-based logic of traditional accounting software. It predicts the next logical number rather than calculating the exact mathematical certainty.

The quiet risk for 50-person finance teams

This update changes how your finance team should handle month-end reporting, but it doesn't replace your financial controller. GPT-5.2 is a probabilistic engine, meaning it guesses the most likely answer based on patterns. Strict financial reconciliation requires deterministic logic, where two plus two equals four every single time.

I see too many business owners assume that because an AI can generate a beautiful cash flow chart, it has accurately checked the underlying ledger. That's a dangerous assumption. If you run a £10M manufacturing business, you can use GPT-5.2 to merge vendor lists, standardise currency formats, or spot anomalies in travel expenses. It's brilliant at the messy data preparation that usually eats up hours of administrative time.

But when it comes to matching bank feeds to invoices, the model will occasionally invent a matching figure to make the prompt resolve smoothly. It lacks the architectural constraints to say a ledger doesn't balance. Your finance team should use this tool to clean the raw materials, but they must run the final reconciliation through deterministic software like Xero or Sage.

Three things to check in your finance stack

1. Audit your current AI usage. Ask your finance team if they are uploading raw bank statements into ChatGPT to categorise expenses. If they are, you need to verify those categorisations against your actual accounting software.
2. Build a data-cleaning workflow. You can safely instruct your team to run messy supplier data through GPT-5.2. Tell them to use the model to standardise dates, fix formatting errors, and remove duplicates before importing the clean CSV into your main database.
3. Watch for accounting vendor updates. If OpenAI continues to improve its native data handling, expect your existing bookkeeping platforms to integrate these specific data-cleaning features directly into their interfaces by Q3. Wait for those native integrations before changing your core financial processes.