---
title: "1,729 vs. 1729 — paper by Reddy et al. (2026)"
description: "LLMs handle arithmetic well when numbers use standard Hindu–Arabic (HA) digits and Western formatting. 21 references."
authors: ["Varshini Reddy", "Craig W Schmidt", "Seth Ebner", "Adam Wiemerslage", "Yuval Pinter", "Chris Tanner"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.518/"
doi: "10.18653/v1/2026.findings-acl.518"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/reddy-2026-1-729-vs-1729-effect/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/reddy-2026-1-729-vs-1729-effect/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/19/) / 1,729 vs. 1729: The Effect of Scripts and Formats on LLM Numeracy

# 1,729 vs. 1729: The Effect of Scripts and Formats on LLM Numeracy

**Authors:** [Varshini Reddy](https://md.research.elamur.ai/free/en/authors/varshini-reddy/), [Craig W Schmidt](https://md.research.elamur.ai/free/en/authors/craig-w-schmidt/), [Seth Ebner](https://md.research.elamur.ai/free/en/authors/seth-ebner/), [Adam Wiemerslage](https://md.research.elamur.ai/free/en/authors/adam-wiemerslage/), [Yuval Pinter](https://md.research.elamur.ai/free/en/authors/yuval-pinter/) and [Chris Tanner](https://md.research.elamur.ai/free/en/authors/chris-tanner/) · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: ACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.518/) · [PDF](https://aclanthology.org/2026.findings-acl.518.pdf) · [DOI 10.18653/v1/2026.findings-acl.518](https://doi.org/10.18653/v1/2026.findings-acl.518) · [OpenAlex](https://openalex.org/W7166804653)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
LLMs handle arithmetic well when numbers use standard Hindu–Arabic (HA) digits and Western formatting. This paper asks whether that skill holds when only the way numbers are written changes. The authors built a controlled benchmark of 336 base arithmetic expressions with integer and decimal operands of 1–8 digits. They rendered each expression in two ways: in 20 non-HA numeral scripts, and in HA digits under 6 regional conventions for decimal markers and digit grouping. Nine models were tested, four large and five small, on three tasks: identifying the script, converting numerals to HA, and doing arithmetic.

Large models mostly recognised and converted scripts correctly. Their arithmetic accuracy, however, collapsed on non-HA numerals, and non-standard formats also reduced performance even though the values were identical. Few-shot prompting recovered much of the lost accuracy. The authors conclude that the failures come mainly from missing context for unfamiliar notation rather than weak arithmetic. This matters for multilingual and regional applications, such as Ukrainian number formats.

## Key points
- **Large models evaluated:** Claude Sonnet 4.5, Gemini 2.5 Pro, Llama-3.3 70B and GPT-4o. The smaller open models scored near zero on script identification and conversion for most scripts, so later analyses focus on the large models.
- **Arithmetic drop:** switching operands from HA to other scripts lowered arithmetic accuracy by roughly 66–87% on average. Gemini performed worst.
- **Shared blind spot:** several models mislabelled Osmanya (Somali) as Osage. Script-identification and conversion accuracy were strongly correlated (r = 0.7022, p = 3.95e−13), suggesting conversion errors start with failing to recognise the script.
- **Format results:**
  - Conventions other than F1 (e.g., 922,436.38) were notably harder. The test set includes F4, a thin-space grouping style the paper attributes to Russia and Ukraine.
  - Multiplication and longer operands also lowered accuracy.
- **Prompting effects:**
  - Naming the script in the prompt did not help.
  - Formatting hints helped only marginally, and requiring formatted outputs made results worse.
  - Few-shot examples gave a large, significant gain (effect 1.694, p \< 1e−8).
  - The dataset is planned for release on Hugging Face.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Ben-Gurion University of the Negev](https://md.research.elamur.ai/free/en/universities/ben-gurion-university-of-the-negev/) — University, Beersheba, Israel · 40 works · 3,099 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [🧮 Computational Theory and Mathematics](https://md.research.elamur.ai/free/en/shelves/computer-science/computational-theory-and-mathematics/) · 321 works

🧭 Topics (1): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,894 works

## Cite

Cite as: Varshini Reddy, Craig W Schmidt, Seth Ebner, Adam Wiemerslage, Yuval Pinter and Chris Tanner. 2026. 1,729 vs. 1729: The Effect of Scripts and Formats on LLM Numeracy. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{reddy20261,
  title = {{1,729 vs. 1729: The Effect of Scripts and Formats on LLM Numeracy}},
  author = {Varshini Reddy, Craig W Schmidt, Seth Ebner, Adam Wiemerslage, Yuval Pinter and Chris Tanner},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.518},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.518}
}
```

- [← previous “We Are (Language) Family”: Adapting Transformer models to related minority languages with linguistic data](https://md.research.elamur.ai/free/en/papers/otal-2026-we-are-language-family-adapting/)
- [next → 100,000+ Movie Reviews from Kazakhstan: Russian, Kazakh, and Code-Switched Texts](https://md.research.elamur.ai/free/en/papers/yeshpanov-2026-100-000-movie-reviews-kazakhstan/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/19/)

## References

21 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Tanja Baeumel, Josef Van Genabith, and Simon Ostermann. 2025. The lookahead limitation: Why multi-operand addition is hard for LLMs. In Proceedings of the 8th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 250– 262, Suzhou, China. Association for Computational Lingui… [find](https://scholar.google.com/scholar?q=Tanja+Baeumel%2C+Josef+Van+Genabith%2C+and+Simon+Ostermann%2E+2025%2E+The+lookahead+limitation%3A+Why+multi%2Doperand+addition+is+hard+for+LLMs%2E+In+Proceedings+of+the+8th+BlackboxNLP+Workshop%3A+Analyzing+and+Inter)
2. Douglas Bates, Martin Mächler, Ben Bolker, and Steve Walker. 2015. [Fitting linear mixed-effects models using lme4](https://md.research.elamur.ai/free/en/papers/bates-2015-fitting-linear-mixed-effects-models/). Journal of Statistical Software, 67(1):1– 48.
3. Antara Raaghavi Bhattacharya, Isabel Papadimitriou, Kathryn Davidson, and David Alvarez-Melis. 2025. Investigating the interaction of linguistic and mathematical reasoning in language models using multilingual number puzzles. In Proceedings of the 2025 Conference on Empirical Methods in Natural Lang… [find](https://scholar.google.com/scholar?q=Antara+Raaghavi+Bhattacharya%2C+Isabel+Papadimitriou%2C+Kathryn+Davidson%2C+and+David+Alvarez%2DMelis%2E+2025%2E+Investigating+the+interaction+of+linguistic+and+mathematical+reasoning+in+language+models+using+mul)
4. N. E. Breslow and D. G. Clayton. 1993. Approximate inference in generalized linear mixed models. Journal of the American Statistical Association, 88(421):9–25. [find](https://scholar.google.com/scholar?q=N%2E+E%2E+Breslow+and+D%2E+G%2E+Clayton%2E+1993%2E+Approximate+inference+in+generalized+linear+mixed+models%2E+Journal+of+the+American+Statistical+Association%2C+88%28421%29%3A9%E2%80%9325%2E)
5. Minh Duc Bui, Kyung Eun Park, Goran Glavaš, Fabian David Schmidt, and Katharina Von Der Wense. 2025. On generalization across measurement systems: LLMs entail more test-time compute for underrepresented cultures. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguist… [find](https://scholar.google.com/scholar?q=Minh+Duc+Bui%2C+Kyung+Eun+Park%2C+Goran+Glava%C5%A1%2C+Fabian+David+Schmidt%2C+and+Katharina+Von+Der+Wense%2E+2025%2E+On+generalization+across+measurement+systems%3A+LLMs+entail+more+test%2Dtime+compute+for+underrepresent)
6. Leo Gao, Stella Biderman, Sid Black, Laurence Golding, Travis Hoppe, Charles Foster, Jason Phang, Horace He, Anish Thite, Noa Nabeshima, Shawn Presser, and Connor Leahy. 2020. The pile: An 800gb dataset of diverse text for language modeling. Preprint, arXiv:2101.00027. [find](https://scholar.google.com/scholar?q=Leo+Gao%2C+Stella+Biderman%2C+Sid+Black%2C+Laurence+Golding%2C+Travis+Hoppe%2C+Charles+Foster%2C+Jason+Phang%2C+Horace+He%2C+Anish+Thite%2C+Noa+Nabeshima%2C+Shawn+Presser%2C+and+Connor+Leahy%2E+2020%2E+The+pile%3A+An+800gb+datas)
7. Mayank Jobanputra, Nils Philipp Walter, Maitrey Mehta, Blerta Veseli, Evan Parker Kelly Chapple, Yifan Wang, Sneha Chetani, Ellie Pavlick, Antonio Vergari, and Vera Demberg. 2025. Can llms subtract numbers? Preprint, arXiv:2511.02795. — [arXiv:2511.02795](https://arxiv.org/abs/2511.02795) [find](https://scholar.google.com/scholar?q=Mayank+Jobanputra%2C+Nils+Philipp+Walter%2C+Maitrey+Mehta%2C+Blerta+Veseli%2C+Evan+Parker+Kelly+Chapple%2C+Yifan+Wang%2C+Sneha+Chetani%2C+Ellie+Pavlick%2C+Antonio+Vergari%2C+and+Vera+Demberg%2E+2025%2E+Can+llms+subtract+nu)
8. Linus Kreitner, Paul Hager, Jonathan Mengedoht, Georgios Kaissis, Daniel Rueckert, and Martin J. Menten. 2025. Efficient numeracy in language models through single-token number embeddings. Preprint, arXiv:2510.06824. — [arXiv:2510.06824](https://arxiv.org/abs/2510.06824) [find](https://scholar.google.com/scholar?q=Linus+Kreitner%2C+Paul+Hager%2C+Jonathan+Mengedoht%2C+Georgios+Kaissis%2C+Daniel+Rueckert%2C+and+Martin+J%2E+Menten%2E+2025%2E+Efficient+numeracy+in+language+models+through+single%2Dtoken+number+embeddings%2E+Preprint%2C+a)
9. R. Thomas McCoy, Ellie Pavlick, and Tal Linzen. 2019. [Right for the wrong reasons: Diagnosing syntactic heuristics in natural language inference](https://md.research.elamur.ai/free/en/papers/mccoy-2019-right-wrong-reasons-diagnosing-syntactic/). In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3428– 3448, Florence, Italy. Association for Computation…
10. Rodrigo Nogueira, Zhiying Jiang, and Jimmy Lin. 2021. [Investigating the limitations of transformers with simple arithmetic tasks](https://md.research.elamur.ai/free/en/papers/nogueira-2021-investigating-limitations-transformers-simple-arithmetic/). Preprint, arXiv:2102.13019.
11. Pedro Javier Ortiz Suárez, Benoît Sagot, and Laurent Romary. 2019. Asynchronous Pipeline for Processing Huge Corpora on Medium to Low Resource Infrastructures. [find](https://scholar.google.com/scholar?q=Pedro+Javier+Ortiz+Su%C3%A1rez%2C+Beno%C3%AEt+Sagot%2C+and+Laurent+Romary%2E+2019%2E+Asynchronous+Pipeline+for+Processing+Huge+Corpora+on+Medium+to+Low+Resource+Infrastructures%2E)
12. Guilherme Penedo, Hynek Kydlíček, Vinko Sabolčec, Bettina Messmer, Negar Foroutan, Amir Hossein Kargaran, Colin Raffel, Martin Jaggi, Leandro Von Werra, and Thomas Wolf. 2025. [Fineweb2: One pipeline to scale them all – adapting pre-training Preprint, data processing to every language](https://md.research.elamur.ai/free/en/papers/penedo-2025-fineweb2-one-pipeline-scale-them/). arXiv:2506.209…
13. Jan-Thorsten Peter, David Vilar, Tobias Domhan, Dan Malkin, and Markus Freitag. 2025. [Mind the gap... or not? how translation errors and evaluation details skew multilingual results](https://md.research.elamur.ai/free/en/papers/peter-2026-mind-gap-or-not-how/). Preprint, arXiv:2511.05162.
14. Freda Shi, Mirac Suzgun, Markus Freitag, Xuezhi Wang, Suraj Srivats, Soroush Vosoughi, Hyung Won Chung, Yi Tay, Sebastian Ruder, Denny Zhou, Dipanjan Das, and Jason Wei. 2022. [Language models are multilingual chain-of-thought reasoners](https://md.research.elamur.ai/free/en/papers/shi-2022-language-models-multilingual-chain-thought/). Preprint, arXiv:2210.03057.
15. Aaditya K. Singh and DJ Strouse. 2024. [Tokenization counts: the impact of tokenization on arithmetic in frontier llms](https://md.research.elamur.ai/free/en/papers/singh-2024-tokenization-counts-impact-tokenization-arithmetic/). Preprint, arXiv:2402.14903. 10687 9
16. Ananya Singha, José Cambronero, Sumit Gulwani, Vu Le, and Chris Parnin. 2023. Tabular representation, noisy operators, and impacts on table structure understanding tasks in llms. Preprint, arXiv:2310.10358. as it encourages more deterministic outputs, helping the models reproduce the specified numer… — [arXiv:2310.10358](https://arxiv.org/abs/2310.10358) [find](https://scholar.google.com/scholar?q=Ananya+Singha%2C+Jos%C3%A9+Cambronero%2C+Sumit+Gulwani%2C+Vu+Le%2C+and+Chris+Parnin%2E+2023%2E+Tabular+representation%2C+noisy+operators%2C+and+impacts+on+table+structure+understanding+tasks+in+llms%2E+Preprint%2C+arXiv%3A2310%2E)
17. Model claude-sonnet-4-5 gpt-4o-2024-08-06 gemini-2.5-pro Llama-3.3-70B-Instruct Llama-3.1-8B-Instruct Olmo-3-7B-Instruct bloomz-7b1 gemma-3-4b-it Qwen3-4B-Instruct-2507 [find](https://scholar.google.com/scholar?q=Model+claude%2Dsonnet%2D4%2D5+gpt%2D4o%2D2024%2D08%2D06+gemini%2D2%2E5%2Dpro+Llama%2D3%2E3%2D70B%2DInstruct+Llama%2D3%2E1%2D8B%2DInstruct+Olmo%2D3%2D7B%2DInstruct+bloomz%2D7b1+gemma%2D3%2D4b%2Dit+Qwen3%2D4B%2DInstruct%2D2507)
18. Eric Wallace, Yizhong Wang, Sujian Li, Sameer Singh, and Matt Gardner. 2019. [Do NLP models know numbers? probing numeracy in embeddings](https://md.research.elamur.ai/free/en/papers/wallace-2019-do-nlp-models-know-numbers/). In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing…
19. Maurice Weber, Daniel Fu, Quentin Anthony, Yonatan Oren, Shane Adams, Anton Alexandrov, Xiaozhong Lyu, Huu Nguyen, Xiaozhe Yao, Virginia Adams, Ben Athiwaratkun, Rahul Chalamala, Kezhen Chen, Max Ryabinin, Tri Dao, Percy Liang, Christopher Ré, Irina Rish, and Ce Zhang. 2024. [Redpajama: an open datas…](https://md.research.elamur.ai/free/en/papers/weber-2024-redpajama-open-dataset-training-large/)
20. Bodo Winter. 2013. Linear models and linear mixed effects models in r with linguistic applications. Preprint, arXiv:1308.5499. [find](https://scholar.google.com/scholar?q=Bodo+Winter%2E+2013%2E+Linear+models+and+linear+mixed+effects+models+in+r+with+linguistic+applications%2E+Preprint%2C+arXiv%3A1308%2E5499%2E)
21. Xiang Zhang, Juntai Cao, Jiaqi Wei, Yiwei Xu, and Chenyu You. 2025. Tokenization constraints in llms: A study of symbolic and arithmetic reasoning limits. Preprint, arXiv:2505.14178. [find](https://scholar.google.com/scholar?q=Xiang+Zhang%2C+Juntai+Cao%2C+Jiaqi+Wei%2C+Yiwei+Xu%2C+and+Chenyu+You%2E+2025%2E+Tokenization+constraints+in+llms%3A+A+study+of+symbolic+and+arithmetic+reasoning+limits%2E+Preprint%2C+arXiv%3A2505%2E14178%2E)

*Updated 2 November 2026*