Research

Paper

AI LLM March 04, 2026

Confidence-Calibrated Small-Large Language Model Collaboration for Cost-Efficient Reasoning

Authors

Chuang Zhang, Zizhen Zhu, Yihao Wei, Bing Tian, Junyi Liu, Henan Wang, Xavier Wang, Yaxiao Liu

Abstract

Large language models (LLMs) demonstrate superior reasoning capabilities compared to small language models (SLMs), but incur substantially higher costs. We propose COllaborative REAsoner (COREA), a system that cascades an SLM with an LLM to achieve a balance between accuracy and cost in complex reasoning tasks. COREA first attempts to answer questions using the SLM, which outputs both an answer and a verbalized confidence score. Questions with confidence below a predefined threshold are deferred to the LLM for more accurate resolution. We introduce a reinforcement learning-based training algorithm that aligns the SLM's confidence through an additional confidence calibration reward. Extensive experiments demonstrate that our method jointly improves the SLM's reasoning ability and confidence calibration across diverse datasets and model backbones. Compared to using the LLM alone, COREA reduces cost by 21.5% and 16.8% on out-of-domain math and non-math datasets, respectively, with only an absolute pass@1 drop within 2%.

Metadata

arXiv ID: 2603.03752
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-03-04
Fetched: 2026-03-05 06:06

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.03752v1</id>\n    <title>Confidence-Calibrated Small-Large Language Model Collaboration for Cost-Efficient Reasoning</title>\n    <updated>2026-03-04T05:55:20Z</updated>\n    <link href='https://arxiv.org/abs/2603.03752v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.03752v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Large language models (LLMs) demonstrate superior reasoning capabilities compared to small language models (SLMs), but incur substantially higher costs. We propose COllaborative REAsoner (COREA), a system that cascades an SLM with an LLM to achieve a balance between accuracy and cost in complex reasoning tasks. COREA first attempts to answer questions using the SLM, which outputs both an answer and a verbalized confidence score. Questions with confidence below a predefined threshold are deferred to the LLM for more accurate resolution. We introduce a reinforcement learning-based training algorithm that aligns the SLM's confidence through an additional confidence calibration reward. Extensive experiments demonstrate that our method jointly improves the SLM's reasoning ability and confidence calibration across diverse datasets and model backbones. Compared to using the LLM alone, COREA reduces cost by 21.5% and 16.8% on out-of-domain math and non-math datasets, respectively, with only an absolute pass@1 drop within 2%.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-03-04T05:55:20Z</published>\n    <arxiv:comment>Accepted to EACL 2026 Main Conference</arxiv:comment>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Chuang Zhang</name>\n    </author>\n    <author>\n      <name>Zizhen Zhu</name>\n    </author>\n    <author>\n      <name>Yihao Wei</name>\n    </author>\n    <author>\n      <name>Bing Tian</name>\n    </author>\n    <author>\n      <name>Junyi Liu</name>\n    </author>\n    <author>\n      <name>Henan Wang</name>\n    </author>\n    <author>\n      <name>Xavier Wang</name>\n    </author>\n    <author>\n      <name>Yaxiao Liu</name>\n    </author>\n  </entry>"
}