ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration

Hongjin Su, Shizhe Diao, Ximing Lu, Mingjie Liu, Jiacheng Xu, Xin Dong, Yonggan Fu, Peter Belcak, Hanrong Ye, Hongxu Yin, Yi Dong, Evelina Bakhturina, Tao Yu, Yejin Choi, Jan Kautz, Pavlo Molchanov
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:115934-115956, 2026.

Abstract

Large language models are powerful generalists, yet solving deep and complex problems such as those of the Humanity’s Last Exam (HLE) remains both conceptually challenging and computationally expensive. We show that small orchestrators managing other models and a variety of tools are able to both push the upper bound of intelligence and improve efficiency in solving difficult agentic tasks. We introduce ToolOrchestra, a method for training small orchestrators that coordinate the use of intelligent tools. ToolOrchestra makes explicit use of reinforcement learning with outcome-, efficiency-, and user-preference-aware rewards. Using ToolOrchestra, we produce Orchestrator, an 8B model that achieves higher accuracy at lower cost than previous tool-use agents while aligning with user preferences on which tools are to be used for a given query. On HLE, Orchestrator achieves a score of 37.1%, outperforming GPT-5 (35.1%) while being 2.5x more efficient. On $\tau ^2$-Bench and FRAMES, Orchestrator surpasses GPT-5 by a wide margin while using only about 30% of the cost. Extensive analysis shows that Orchestrator achieves the best trade-off between performance and cost under multiple metrics, and generalizes robustly to previously unseen tools. These results demonstrate that composing diverse tools with a lightweight orchestration model is both more efficient and more effective than existing methods, paving the way for practical and scalable tool-augmented reasoning systems. These results demonstrate that orchestrating diverse tools with lightweight agents is not only more efficient, but also more effective, paving the way for practical and scalable tool-augmented reasoning systems.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-su26d, title = {{T}ool{O}rchestra: Elevating Intelligence via Efficient Model and Tool Orchestration}, author = {Su, Hongjin and Diao, Shizhe and Lu, Ximing and Liu, Mingjie and Xu, Jiacheng and Dong, Xin and Fu, Yonggan and Belcak, Peter and Ye, Hanrong and Yin, Hongxu and Dong, Yi and Bakhturina, Evelina and Yu, Tao and Choi, Yejin and Kautz, Jan and Molchanov, Pavlo}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {115934--115956}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/su26d/su26d.pdf}, url = {https://proceedings.mlr.press/v306/su26d.html}, abstract = {Large language models are powerful generalists, yet solving deep and complex problems such as those of the Humanity’s Last Exam (HLE) remains both conceptually challenging and computationally expensive. We show that small orchestrators managing other models and a variety of tools are able to both push the upper bound of intelligence and improve efficiency in solving difficult agentic tasks. We introduce ToolOrchestra, a method for training small orchestrators that coordinate the use of intelligent tools. ToolOrchestra makes explicit use of reinforcement learning with outcome-, efficiency-, and user-preference-aware rewards. Using ToolOrchestra, we produce Orchestrator, an 8B model that achieves higher accuracy at lower cost than previous tool-use agents while aligning with user preferences on which tools are to be used for a given query. On HLE, Orchestrator achieves a score of 37.1%, outperforming GPT-5 (35.1%) while being 2.5x more efficient. On $\tau ^2$-Bench and FRAMES, Orchestrator surpasses GPT-5 by a wide margin while using only about 30% of the cost. Extensive analysis shows that Orchestrator achieves the best trade-off between performance and cost under multiple metrics, and generalizes robustly to previously unseen tools. These results demonstrate that composing diverse tools with a lightweight orchestration model is both more efficient and more effective than existing methods, paving the way for practical and scalable tool-augmented reasoning systems. These results demonstrate that orchestrating diverse tools with lightweight agents is not only more efficient, but also more effective, paving the way for practical and scalable tool-augmented reasoning systems.} }
Endnote
%0 Conference Paper %T ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration %A Hongjin Su %A Shizhe Diao %A Ximing Lu %A Mingjie Liu %A Jiacheng Xu %A Xin Dong %A Yonggan Fu %A Peter Belcak %A Hanrong Ye %A Hongxu Yin %A Yi Dong %A Evelina Bakhturina %A Tao Yu %A Yejin Choi %A Jan Kautz %A Pavlo Molchanov %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-su26d %I PMLR %P 115934--115956 %U https://proceedings.mlr.press/v306/su26d.html %V 306 %X Large language models are powerful generalists, yet solving deep and complex problems such as those of the Humanity’s Last Exam (HLE) remains both conceptually challenging and computationally expensive. We show that small orchestrators managing other models and a variety of tools are able to both push the upper bound of intelligence and improve efficiency in solving difficult agentic tasks. We introduce ToolOrchestra, a method for training small orchestrators that coordinate the use of intelligent tools. ToolOrchestra makes explicit use of reinforcement learning with outcome-, efficiency-, and user-preference-aware rewards. Using ToolOrchestra, we produce Orchestrator, an 8B model that achieves higher accuracy at lower cost than previous tool-use agents while aligning with user preferences on which tools are to be used for a given query. On HLE, Orchestrator achieves a score of 37.1%, outperforming GPT-5 (35.1%) while being 2.5x more efficient. On $\tau ^2$-Bench and FRAMES, Orchestrator surpasses GPT-5 by a wide margin while using only about 30% of the cost. Extensive analysis shows that Orchestrator achieves the best trade-off between performance and cost under multiple metrics, and generalizes robustly to previously unseen tools. These results demonstrate that composing diverse tools with a lightweight orchestration model is both more efficient and more effective than existing methods, paving the way for practical and scalable tool-augmented reasoning systems. These results demonstrate that orchestrating diverse tools with lightweight agents is not only more efficient, but also more effective, paving the way for practical and scalable tool-augmented reasoning systems.
APA
Su, H., Diao, S., Lu, X., Liu, M., Xu, J., Dong, X., Fu, Y., Belcak, P., Ye, H., Yin, H., Dong, Y., Bakhturina, E., Yu, T., Choi, Y., Kautz, J. & Molchanov, P.. (2026). ToolOrchestra: Elevating Intelligence via Efficient Model and Tool Orchestration. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:115934-115956 Available from https://proceedings.mlr.press/v306/su26d.html.

Related Material