I was curious enough to have Codex create a similar benchmark: https://github.com/jcheng5/table-formats

With 1000 rows and 100 samples and markdown-kv, I got these scores:

- gpt-4.1-nano: 52%

- gpt-4.1-mini: 72%

- gpt-4.1: 93%

- gpt-5: 100%

I was so surprised by gpt-5 getting 100% that I ran it again with 1000 samples. It got 999 correct, and one wrong.

To reproduce it yourself, clone the repo, add a .env file with OPENAI_API_KEY, `uv sync`, and then run:

    uv run inspect eval evals/table_formats_eval.py@table_formats_markdown_kv --model openai/gpt-5 --limit 100

Update: Also, number of rows makes a massive difference, unsurprisingly; at 100 rows, gpt-4.1-nano scores 95%+ for both markdown-kv and csv. Both model and record count seem to matter a lot more than format.

    uv run inspect eval evals/table_formats_eval.py@table_formats_csv --model openai/gpt-5 --limit 100
    uv run inspect eval evals/table_formats_eval.py@table_formats_json --model openai/gpt-5 --limit 100

    uv add google-genai
    uv run scripts/run_benchmarks.py --models google/gemini-2.5-pro --formats markdown_kv --limit 100

> where accuracy is paramount

> accuracy: 60%

Not to mention that the least poorly performing format is probably the stupidest way to encode tabular data, beating even XML. But I guess that’s the new normal because we’re trying to shoehorn conversational AI models to every use case rather than, say, training finetunes that are better at particular tasks. (Yes, of course you can’t train finetunes when the model is a proprietary black box on someone else’s computer.) Something about hammers and nails…

    CSV: 84.25%
    Markdown Table: 82.65%
    YAML: 81.85%
    JSON Lines (jsonl): 79.85%
    Markdown key-value: 79.83%
    Pipe-delimited: 79.45%
    Natural language summary: 78.65%
    JSON: 77.73%
    HTML table: 75.80%
    XML: 73.80%

    - changing the order of the data on each table type
    - changing the order of the questions

    Where do you eat?
    A) floor
    B) table
    C) dirt

  ## Record 1
  
  ```
  id: 1
  name: Charlie A0
  age: 56
  city: New York
  department: Operations
  salary: 67896
  years_experience: 7
  project_count: 1
  ```

Title says "LLMs" (plural) but they only tested one

> We only tested OpenAI’s GPT-4.1 nano.

The test really needed to be run on multiple data sizes (50, 100, 500, 1000, 5000). The more token efficient formats would probably eventually overtake the token heavy ones due to context pollution. All this test really says is what performs best for 1 particular model at one particular context length.

Interesting. Curious to reproduce across models, I made a comprehensive eval based on your post and ran it against 30 models, each tasked with recalling specific data from 500 rows in different tabular formats. Have a look at the results here: https://weval.org/analysis/table-format-sensitivity__combine...

As you can see it's near 100% recall across all formats for a good chunk of frontier models, with a few (curiously, mostly Claude) failing at basic prompt adherance ("Return just the number") but still returning the right answers. The major failures are from Mistral Medium, Llama Maverick, Llama 3 70b Instruct, Mistral Nemo, Gemma 3 12b It, GPT 4o/4.1 Mini etc.

Based on these limited tests, here's the leaderboards on formats FWIW:

    CSV: 84.25%
    Markdown Table: 82.65%
    YAML: 81.85%
    JSON Lines (jsonl): 79.85%
    Markdown key-value: 79.83%
    Pipe-delimited: 79.45%
    Natural language summary: 78.65%
    JSON: 77.73%
    HTML table: 75.80%
    XML: 73.80%

IMO the biggest takeaway really is: Use the best model you can reasonably afford, then the format chosen will matter less. The cheapest 100%-coverage models are Gemini 2.5 Flash and Deepseek Chat V3.1 FWIW. However, if you have no control over model, then use CSV or Markdown Table as these have highest chance of success.

The MAJOR issue that we might not want to admit is that there are a thousand confounders that prevent any meaningful canonical learning here. Crucially: The data within the tabular structure itself matters HUGELY. The scary probabilistic nature of LLMs mean the very subject of your queries can affect how the query is run, which is quite absurd from a IO/computing purity perspective. This is why tooling is so important. Enable the LLM to write and execute code safely, and you don't need to worry about such free-prose frailties.

Markdown being the most understandable format lines up with my anecdotal experience. But I also was only using OpenAI models at the time. I do think there are a lot of unexplored methods/table transformation tools that we haven't even thought of yet.

Bizarre conclusions when on average all the formats perform poorly with average accuracy of 50%. Sure 60% is better than 40% but they are both unusable if you actually care about numbers...

I wonder how this compares to a more agentic approach where the LLM composes SQL queries to answer the questions, for example.

Can someone explain why one would want to use an LLM to read tabular data? This is something even trivial code could do while using far fewer compute and energy resources.

Great benchmark! It highlights an important but often downstream problem. In real-world pipelines, the bigger issue comes before this: extracting tables from PDFs or scans without breaking their layout. Once the structure is lost (merged headers, nested cells, footnotes, etc.), no data format can fully recover it.

Check out LLMWhisperer from Unstract —> it preserves table and layout fidelity when converting documents for LLM use. You can try it on complex PDFs or forms here: https://pg.llmwhisperer.unstract.com (no signup needed)

Layout preservation upstream often improves downstream accuracy more than choosing between CSV, JSON, or Markdown. Find more details here: https://unstract.com/llmwhisperer/

The article has interesting data. But it’s frustrating to read AI generated text like this:

> Performance Optimization: Reducing processing overhead while maintaining accuracy

What on earth does it mean that this “optimized performance”? This is nonsensical content. Performance wasn’t even measured, accuracy was. You can tell this was AI generated because “ Reducing processing overhead while maintaining accuracy” would likely be true for a perf optimization, but it has no meaning whatsoever in the context of the article.

This really throws into question whether I can take the rest of the article and data seriously.

This is a bit silly way to use LLMs to process tabular data. In reality, you'd ask it to write functions and execute them. First you'd ask it to create a type definition from the table, then ask it to create functions to process the data.

"Write a function to find years of experience by name? Return just the number, e.g. '12'."

It works much better, and it can single-shot many of the processing requirements just from type definitions it can infer from the data.

This way it's easier to stick to tabular formats that have easy reading libraries, like with TypeScript/JavaScript JSON, and with Python, maybe CSV...

Tbh I am more interested in processing data and formatting it to tabular forms than extracting data from tabular forms. One of the main uses I see in LLMs is structuring unstructured/semistructured data. I may occasionally feed a table to an LLM and ask such kinds of questions when I feel lazy, but I see no serious application of this as compared with using whatever language/library to process the data from the table (whether using an llm or not in the whole process). The point of having structured data is exactly this. But much more often I feed data to an llm and ask it to create a table.

Curious how text-aligned tabular formats work for LLMs considering humans probably find them more readable than other formats

                                                                 System Sales(a)  
                                           Number of Units         (in Millions)  
         ──────────────────────────────────────────────────────────────────────── 
          KFC Division                              31,981    $           34,452  
          Taco Bell Division                         8,757                17,193  
          Pizza Hut Division                        20,225                13,108  
          Habit Burger & Grill Division                383                   713  
          YUM                                       61,346    $           65,466  

I'm seeing pretty good success with extracting data out of 10-Qs which are formatted like this by default using the `edgartools` library's default `filing.text()` method.

They don’t understand any table formats; as shown by these results.

They can transform information in tables but information is lost due to that lack of understanding.

Really interesting post. I ran into some of the limitations of working with tables and LLM's last year.

I experimented with an approach to use the llm to generate a bespoke transformation machine that uses an LLM to generate a series of transform steps to extracting key data from large data sets.

https://tombers.github.io/oblique-angles/ai/education/2025/0...

The current OCR approach typically relies on a Vision-Language Model (VLM) to convert a table into a JSON structure. However, a table inherently has a 2D spatial structure, while Large Language Models (LLMs) are optimized for processing 1D sequential text. This creates a fundamental mismatch between the data representation and the model’s input format.

Most existing pipelines address this by preprocessing the table into a linearized 1D string before passing it to the LLM — a question-agnostic step that may lose structural information.

Instead, one could retain the original table form and, when a question is asked, feed both the question and the original table (as an image) directly into the VLM. This approach allows the model to reason over the data in its native 2D domain, providing a more natural and potentially more accurate solution.

These sort of experiments and results are really important for language model implementation. This has a tangible implication for my AI startup and how we approach tool design.

Much more important than citation farming a paper on 1 % improved performance

Inputs were not long enough to properly see either of the true wins in terms of reduced token counts for terser formats or their benefits in terms of avoiding stuffing the context window thereby potentially reducing accuracy. The test really needs to be conducted across multiple dimensions!

It appears that this is just testing data retrieval from somewhere in the table? Do the results translate to something where data analysis is performed? From something as simple as summing across rows or averages to generating graphs.

I once tried to get Claude and ChatGPT to build me a excel financial model, failed pretty hard. The models seem to lose track where they are in a table

This article screams for a accuracy vs. tokens plot. Thanks though, interesting results.

I am not an expert on the subject but i suggest that you can also save context space by using shorter XML element names (like f instead of function, c instead of class, etc.). Just add a legend at the top or bottom to explain what each abbreviation means, LLMs can figure out the mapping without issues. I use this approach when generating project structure maps with Tree-sitter. I did a quick comparison and didn't notice much degradation with claude, so the context space you save may make it worthwhile. I would be interested to see a proper comparison.

Only testing GPT-4.1-nano makes this basically useless. Most people are almost certainly using GPT-5 mini or better. This very poor analysis is like an LLM literacy test for readers.

This is an interesting theoretical exercise but please for the love of god don't actually use an LLM to search tabular data. This is a solved problem. Free software does this with 100% accuracy and insane efficiency.

We ended up making middleware for LLM 'tools/functions' that take common data/table formats like CSV, Excel and JSON.

The tool uses an LLM to write code to parse the data and conduct the analysis to return back to the LLM. Otherwise, we found pumping raw table data into a LLM is just not reliable, even if you go to the effort to conduct analysis on smaller chunks and merge the results.

KSON? (I'm a complete ignoramus in this area but recently read about KSON in a piece posted here at HN.)

https://ochagavia.nl/blog/configuration-files-are-user-inter...

https://news.ycombinator.com/item?id=45291858 (135 comments)

> 60.7%

Why would anyone trust the output of an LLM, if it is barely better than guessing and much much worse than humans?

GPT-5 shows more impressive numbers, but for that particular task, the precision should be 100% - always. No matter how large the data set is or in which format. Why are we doing this?

I've found that xml is surprisingly good for llms when it comes to table extraction in production. I only found out when I send the raw xml storage format to benchmark again various flavours of everything else. XML turns out to the best format for tables that have more than three levels of nesting.

That's a cool concept - would be curious about a more common setup for agentic data analysis (ex: for using in Claude Code) like:

* Multiple tasks vs 1

* O3/o3-mini + 4o/4o-mini instead of nano

* Extra credit: Inside a fixed cost/length reasoning loop

Ex: does the md-kv benefit disappear with smarter models that you'r typically use, and thus just become a 2-3x cost?

Was a bit surprised about the low csv performance, in my exp. it‘s very good (use it a lot with Excel and small tables, well below 100 rows).

As markdown kv performs so well, I am now curious about TOML.

I find this extremely surprising. I would have expected dict structures to have higher semantic context associated with them.

Hmmm. I’ve been using YAML data for tables for a while now, and had pretty good results.

Super surprised, I would expect CSV to beat all the others. And Markdown KV is something I hear first time about.

I’d be interested in testing different data formats when using the structured outputs api

I'm surprised by the accuracy, in practice, I feel like I generally have a lot better results

There are other studies on this topic with similar results across LLM systems:

Y. Sui, M. Zhou, M. Zhou, S. Han, and D. Zhang, “Table Meets LLM: Can Large Language Models Understand Structured Table Data? A Benchmark and Empirical Study,” in Proceedings of the 17th ACM International Conference on Web Search and Data Mining, Merida Mexico: ACM, Mar. 2024, pp. 645–654. doi: 10.1145/3616855.3635752.

C. Pang, Y. Cao, C. Yang, and P. Luo, “Uncovering Limitations of Large Language Models in Information Seeking from Tables,” June 06, 2024, arXiv: arXiv:2406.04113. doi: 10.48550/arXiv.2406.04113.

Misleading title, just one LLM was tested.

interesting. I'm curious how this compares across different model families.

Great idea. Very limited execution. If they release the source data and question set, I'll repeat with more LLMs to flesh out the findings.

accuracy: 60%

This should have been a python script.

How much of the current peak of the Gartner Hype Cycle should just be python scripts?

In mice.

Or in this case gpt-4.1-nano

maybe be org table