2026-01-08
From 1997 to 2025, there have been 13,947 opinions published by the Tax Court. A total of 166 of these opinions (1.2%) were more than 80 pages long. Below is a histogram of the number of pages for the Tax Court opinions that were 80 pages or shorter. The average length of the opinions is 18 pages.
I have been experimenting with running some large language models (LLMs) locally on my computer. Assuming 4 characters per token, the average Tax Court opinion has about 6K tokens. The gemma3:4b LLM has a context length of up to 128K tokens. Therefore, gemma3:4b should be able to analyze nearly all Tax Court opinions. Tax Court opinions with 80 pages average about 27K tokens. (Apparently smaller LLMs, such as those with 4 billion parameters, are now being referred to as small language models, or SLMs.)
So far I have been impressed with the summarization capabilities of gemma3:4b. However, I still find LLMs wholly unreliable for tax research.
Tags: Statistics, Python, Other - Artificial Intelligence (AI)
The posts on this blog have not been verified for accuracy. You should consult an attorney for legal advice regarding your own situation. These posts are not updated for changes in the tax laws. Further, these posts should not be relied upon for any purpose whatsoever.
Copyright © 2004 – 2026, Andrew Mitchel LLC. All rights reserved.