aiqbench analysis
This research reveals that standard token-level entropy fails as a confidence metric for Small Language Models (SLMs), effectively rendering them "blind" to their own uncertainty in most cases. Instead, the study demonstrates that semantic entropy—derived from sampling and clustering—provides a viable signal for routing queries to larger expert models, yielding significant accuracy gains without relying on architectural compatibility. The core insight is that optimizing SLM performance involves intelligent compute allocation rather than simple computational savings, fundamentally shifting how uncertainty is managed at scale.
Do small language models know what they don't know?
arXiv:2609.20824v1 Announce Type: new Abstract: We explore whether entropy-based confidence signals can be leveraged to improve the accuracy of Small Language Models (SLMs) with fewer than 3 billion parameters, running entirely on consumer hardware. We evaluate seven distinct approaches, including token-level entropy early stopping, semantic entropy estimation, and uncertainty-aware routing to larger expert models, across 7 model pairs and 5 standard NLU benchmarks. Our key finding is that token-level entropy is effectively blind in SLMs: in 91% of dataset-model combinations, mean token entropy is near zero regardless of answer correctness, rendering token-based confidence signals unusable at this scale. We demonstrate that semantic entropy, computed by generating multiple samples, clustering answers by meaning, and measuring distributional uncertainty, recovers a viable confidence signal. Using semantic entropy to selectively route uncertain queries to a larger expert model yields accuracy improvements of up to +50 percentage points. Notably, cross-family routing (e.g., SmolLM 360M to Phi-3.5-mini) averages +22.0% improvement compared to +6.8% for same-family routing, revealing that expert model quality matters more than architectural compatibility. Our results suggest that the value proposition for entropy-based methods in SLMs is not computational savings but intelligent compute allocation: spending more tokens where they matter most.
Title:Do small language models know what they don't know?
Abstract:We explore whether entropy-based confidence signals can be leveraged to improve the accuracy of Small Language Models (SLMs) with fewer than 3 billion parameters, running entirely on consumer hardware. We evaluate seven distinct approaches, including token-level entropy early stopping, semantic entropy estimation, and uncertainty-aware routing to larger expert models, across 7 model pairs and 5 standard NLU benchmarks. Our key finding is that token-level entropy is effectively blind in SLMs: in 91% of dataset-model combinations, mean token entropy is near zero regardless of answer correctness, rendering token-based confidence signals unusable at this scale. We demonstrate that semantic entropy, computed by generating multiple samples, clustering answers by meaning, and measuring distributional uncertainty, recovers a viable confidence signal. Using semantic entropy to selectively route uncertain queries to a larger expert model yields accuracy improvements of up to +50 percentage points. Notably, cross-family routing (e.g., SmolLM 360M to Phi-3.5-mini) averages +22.0% improvement compared to +6.8% for same-family routing, revealing that expert model quality matters more than architectural compatibility. Our results suggest that the value proposition for entropy-based methods in SLMs is not computational savings but intelligent compute allocation: spending more tokens where they matter most.
Computation and Language (cs.CL)
arXiv:2609.20824 [cs.CL]
(or arXiv:2609.20824v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.20824 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 294 words
Related coverage
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space · Oct 13, 2026
Humanity’s Last Invention — Richard Socher of Recursive
Latent Space · Sep 14, 2026
Podcast: The Future of AI: from Enterprise Adoption to Open Source Sovereignty
InfoQ · Sep 25, 2026