---
title: "What Moves the Pareto Frontier in Tool-Using Agents? A"
description: "The work offers a reusable protocol and budget-specific guidance for choosing agent designs. 18 references."
authors: ["Rishi N. Simhadri"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)"
source: "https://aclanthology.org/2026.acl-srw.82/"
doi: "10.18653/v1/2026.acl-srw.82"
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/simhadri-2026-what-moves-pareto-frontier-tool/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/simhadri-2026-what-moves-pareto-frontier-tool/)

[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/47/) / What Moves the Pareto Frontier in Tool-Using Agents? A Compute-Aware Study of ReAct Variants

# What Moves the Pareto Frontier in Tool-Using Agents? A Compute-Aware Study of ReAct Variants

**Authors:** Rishi N. Simhadri · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/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.acl-srw.82/) · [PDF](https://aclanthology.org/2026.acl-srw.82.pdf) · [DOI 10.18653/v1/2026.acl-srw.82](https://doi.org/10.18653/v1/2026.acl-srw.82) · [OpenAlex](https://openalex.org/W7166800698)

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

## Summary
LLM agents that call tools are usually ranked by accuracy alone, even though real deployments must stay within inference budgets. This student workshop paper examines whether popular ways of strengthening ReAct-style agents still look worthwhile once token spending is counted. Using Gemini 2.5 Flash, the author tests five configurations on two tool-heavy benchmarks, HotPotQA and FEVER, and on GSM8K as a math-focused contrast. The configurations are plain ReAct, CoT-SC with LLM-based synthesis, majority voting, Reflexion and TCAR, a one-pass post-hoc revision step. Per-instance token logs are used to build cumulative accuracy-versus-budget curves, Pareto frontiers and marginal-cost views after the runs. Reflexion is most accurate on the tool-heavy tasks, but its cost and accuracy are optimistic because retries stop when the answer matches the gold label. Sampling methods spend several times more tokens for small improvements there. On GSM8K, CoT-SC leads and revision helps far more. The work offers a reusable protocol and budget-specific guidance for choosing agent designs.

## Key points
- Each configuration was run with three random seeds. HotPotQA and FEVER used 500 instances per seed, and GSM8K used 1,015.
- CoT-SC and majority voting draw n = 3 trajectories, and Reflexion allows up to 7 trials. The variants are adaptations inspired by the original methods rather than exact reproductions.
- On the tool-heavy benchmarks, sampling-based methods used roughly 3–5× more tokens than baseline ReAct while gaining relatively little accuracy.
- On GSM8K, CoT-SC gained about +12 pp and TCAR +10.2 pp, the latter at only 1.28× ReAct's cost. On HotPotQA and FEVER, TCAR added just +0.9–1.6 pp.
- The author attributes the efficiency gap mainly to repeating full tool trajectories. Stated limitations include reliance on a single base model, too few seeds for significance testing, and an offline evaluation of Reflexion that uses gold-label stopping.

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

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

## Cite

Cite as: Rishi N. Simhadri. 2026. What Moves the Pareto Frontier in Tool-Using Agents? A Compute-Aware Study of ReAct Variants. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop).

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

```bibtex
@inproceedings{simhadri2026what,
  title = {{What Moves the Pareto Frontier in Tool-Using Agents? A Compute-Aware Study of ReAct Variants}},
  author = {Rishi N. Simhadri},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-srw.82},
  url = {https://doi.org/10.18653/v1/2026.acl-srw.82}
}
```

- [← previous What Makes AI Research Replicable? Executable Knowledge Graphs as Scientific Knowledge Representations](https://md.research.elamur.ai/free/en/papers/luo-2026-what-makes-ai-research-replicable/)
- [next → What Question Did You Answer? Refining Contact Center Evaluation Plans via Backward Questions](https://md.research.elamur.ai/free/en/papers/sood-2026-what-question-did-you-answer/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/47/)

## References

18 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Lingjiao Chen, Matei Zaharia, and James Zou. 2024. [FrugalGPT: How to use large language models while reducing cost and improving performance](https://md.research.elamur.ai/free/en/papers/chen-2023-frugalgpt-use-large-language-models/). Transactions on Machine Learning Research (TMLR).
2. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. [Training verifiers to solve math word problems](https://md.research.elamur.ai/free/en/papers/cobbe-2021-training-verifiers-solve-math-word/). arXiv preprint arXiv:2110.14168.
3. Mehmet Hamza Erol, Batu El, Mirac Suzgun, Mert Yuksekgonul, and James Zou. 2025. Cost-of-pass: An economic framework for evaluating language models. Preprint, arXiv:2504.13359. — [arXiv:2504.13359](https://arxiv.org/abs/2504.13359) [find](https://scholar.google.com/scholar?q=Mehmet+Hamza+Erol%2C+Batu+El%2C+Mirac+Suzgun%2C+Mert+Yuksekgonul%2C+and+James+Zou%2E+2025%2E+Cost%2Dof%2Dpass%3A+An+economic+framework+for+evaluating+language+models%2E+Preprint%2C+arXiv%3A2504%2E13359%2E+%E2%80%94+%5BarXiv%3A2504%2E13359%5D%28ht)
4. Luyu Gao, Zhuyun Dai, Panupong Pasupat, Anthony Chen, Yoon Kim, Shankar Natarajan, Nanyun Peng, Tao Yu, and 1 others. 2023. [Rarr: Researching and revising what language models say, using language models](https://md.research.elamur.ai/free/en/papers/gao-2023-rarr-researching-revising-what-language/). In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL)…
5. Kaisa Miettinen. 1999. Nonlinear Multiobjective Optimization. Springer. [find](https://scholar.google.com/scholar?q=Kaisa+Miettinen%2E+1999%2E+Nonlinear+Multiobjective+Optimization%2E+Springer%2E)
6. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. [Reflexion: Language agents with verbal reinforcement learning](https://md.research.elamur.ai/free/en/papers/shinn-2023-reflexion-language-agents-verbal-reinforcement/). In Advances in Neural Information Processing Systems (NeurIPS).
7. Charlie Snell, Jaehoon Lee, Kelvin Xu, and Aviral Kumar. 2024. [Scaling llm test-time compute optimally can be more effective than scaling model parameters](https://md.research.elamur.ai/free/en/papers/snell-2024-scaling-llm-test-time-compute/). Preprint, arXiv:2408.03314.
8. Junlin Wang, Siddhartha Jain, Dejiao Zhang, Baishakhi Ray, Varun Kumar, and Ben Athiwaratkun. 2024. Reasoning in token economies: Budget-aware evaluation of LLM reasoning strategies. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 19916– 19939… [find](https://scholar.google.com/scholar?q=Junlin+Wang%2C+Siddhartha+Jain%2C+Dejiao+Zhang%2C+Baishakhi+Ray%2C+Varun+Kumar%2C+and+Ben+Athiwaratkun%2E+2024%2E+Reasoning+in+token+economies%3A+Budget%2Daware+evaluation+of+LLM+reasoning+strategies%2E+In+Proceedings+of)
9. Google. 2025. Gemini 2.5 flash model documentation. [find](https://scholar.google.com/scholar?q=Google%2E+2025%2E+Gemini+2%2E5+flash+model+documentation%2E)
10. Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023. [Self-consistency improves chain of thought reasoning in language models](https://md.research.elamur.ai/free/en/papers/wang-2022-self-consistency-improves-chain-thought/). In International Conference on Learning Representations (ICLR).
11. Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. 2024. [SWE-bench: Can language models resolve real-world GitHub issues?](https://md.research.elamur.ai/free/en/papers/jimenez-2024-swe-bench-can-language-models/) In International Conference on Learning Representations (ICLR).
12. Yangzhen Wu, Zhiqing Sun, Shanda Li, Sean Welleck, and Yiming Yang. 2025. [Inference scaling laws: An empirical analysis of compute-optimal inference for problem-solving with language models](https://md.research.elamur.ai/free/en/papers/wu-2024-inference-scaling-laws-empirical-analysis/). In International Conference on Learning Representations (ICLR).
13. Sayash Kapoor, Benedikt Stroebl, Zachary S. Siegel, Nitya Nadgir, and Arvind Narayanan. 2025. AI agents that matter. Transactions on Machine Learning Research (TMLR). [find](https://scholar.google.com/scholar?q=Sayash+Kapoor%2C+Benedikt+Stroebl%2C+Zachary+S%2E+Siegel%2C+Nitya+Nadgir%2C+and+Arvind+Narayanan%2E+2025%2E+AI+agents+that+matter%2E+Transactions+on+Machine+Learning+Research+%28TMLR%29%2E)
14. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik Narasimhan. 2023a. Tree of thoughts: Deliberate problem solving with large language models. In Advances in Neural Information Processing Systems (NeurIPS). [find](https://scholar.google.com/scholar?q=Shunyu+Yao%2C+Dian+Yu%2C+Jeffrey+Zhao%2C+Izhak+Shafran%2C+Thomas+L%2E+Griffiths%2C+Yuan+Cao%2C+and+Karthik+Narasimhan%2E+2023a%2E+Tree+of+thoughts%3A+Deliberate+problem+solving+with+large+language+models%2E+In+Advances+in+)
15. Tengxiao Liu, Zifeng Wang, Jin Miao, I-Hung Hsu, Jun Yan, Jiefeng Chen, Rujun Han, Fangyuan Xu, Yanfei Chen, Ke Jiang, Samira Daruki, Yi Liang, William Yang Wang, Tomas Pfister, and Chen-Yu Lee. 2025. Budget-aware tool-use enables effective agent scaling. Preprint, arXiv:2511.17006. — [arXiv:2511.17006](https://arxiv.org/abs/2511.17006) [find](https://scholar.google.com/scholar?q=Tengxiao+Liu%2C+Zifeng+Wang%2C+Jin+Miao%2C+I%2DHung+Hsu%2C+Jun+Yan%2C+Jiefeng+Chen%2C+Rujun+Han%2C+Fangyuan+Xu%2C+Yanfei+Chen%2C+Ke+Jiang%2C+Samira+Daruki%2C+Yi+Liang%2C+William+Yang+Wang%2C+Tomas+Pfister%2C+and+Chen%2DYu+Lee%2E+2025%2E)
16. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023b. [React: Synergizing reasoning and acting in language models](https://md.research.elamur.ai/free/en/papers/yao-2022-react-synergizing-reasoning-acting-language/). In International Conference on Learning Representations (ICLR).
17. Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, Shudan Zhang, Xiang Deng, Aohan Zeng, Zhengxiao Du, Chenhui Zhang, Sheng Shen, Tianjun Zhang, Yu Su, Huan Sun, and 3 others. 2024. [AgentBench: Evaluating LLMs as agents](https://md.research.elamur.ai/free/en/papers/liu-2023-agentbench-evaluating-llms-agents/). In International…
18. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-refine: Iterative refinement with… [find](https://scholar.google.com/scholar?q=Aman+Madaan%2C+Niket+Tandon%2C+Prakhar+Gupta%2C+Skyler+Hallinan%2C+Luyu+Gao%2C+Sarah+Wiegreffe%2C+Uri+Alon%2C+Nouha+Dziri%2C+Shrimai+Prabhumoye%2C+Yiming+Yang%2C+Shashank+Gupta%2C+Bodhisattwa+Prasad+Majumder%2C+Katherine+Her)

*Updated 1 November 2026*