# Routes of Reasoning: Reading and Reshaping Test-Time Compute in MoE Models

DevFeed: [Routes of Reasoning: Reading and Reshaping Test-Time Compute in MoE Models](<https://devfeed.tech/articles/routes-of-reasoning-reading-and-reshaping-test-time-compute-in-moe-models-118403.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.05919>)

Author: Shengjie Wang, Mohsen Fayyaz, Nanyun Peng

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>)

Tags: [arxiv](<https://devfeed.tech/tags/arxiv.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [behaviors](<https://devfeed.tech/tags/behaviors.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [code](<https://devfeed.tech/tags/code.md>), [compute](<https://devfeed.tech/tags/compute.md>), [conclusion](<https://devfeed.tech/tags/conclusion.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [domains](<https://devfeed.tech/tags/domains.md>), [execution](<https://devfeed.tech/tags/execution.md>), [experts](<https://devfeed.tech/tags/experts.md>), [features](<https://devfeed.tech/tags/features.md>), [hidden](<https://devfeed.tech/tags/hidden.md>), [inference](<https://devfeed.tech/tags/inference.md>), [models](<https://devfeed.tech/tags/models.md>), [moe](<https://devfeed.tech/tags/moe.md>), [reading](<https://devfeed.tech/tags/reading.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>)

## AI overview

The paper uses Mixture-of-Experts router signals to identify reasoning behaviors and selectively reduce overthinking during inference. Sentence-level trace segments are labeled with six behaviors; linear probes on router logits predict those labels nearly as well as probes on the full residual stream using about half as many features. Behavior-associated experts generalize across six datasets and five domains. Editing router scores at inference reduced generated tokens by a median of 30.4% (up to 63.3%) across four models and four benchmarks. Reported accuracy fell by no more than 0.5 percentage points and often improved; the intervention was designed to encourage stopping after solving a problem without skipping necessary reasoning.

## Source excerpt

arXiv:2611.05919v1 Announce Type: new Abstract: Reasoning models allocate test-time compute across distinct behaviors such as planning, deriving, and verifying. However, models frequently "overthink," continuing to verify and re-derive long after finding a correct answer. This wastes expensive compute and can even degrade performance. Existing interventions, such as token budgets or steering, primarily regulate reasoning duration rather than targeting specific behaviors. In this work, we demonstrate that Mixture-of-Experts (MoE) routers provide a natural, training-free interface for behavioral control. We first split reasoning traces into sentence-level segments and label each with one of six reasoning behaviors (planning, execution, restatement, fact, reflection, and conclusion), so that every token carries the behavior of its segment. Linear probes on router logits predict this label nearly as well as probes on the residual stream, the model's full hidden state and the standard target for probing, while using about half as many features. These signals generalize: experts associated with a reasoning behavior on one dataset predict the same behavior on the others, across six datasets spanning five domains (math, code, science, knowledge, and commonsense). Building on this, we selectively edit expert router scores during inference to curb overthinking. Across four models and four benchmarks, this intervention reduces the number of generated tokens by a median of 30.4% (and up to 63.3%). Crucially, it encourages the model to stop once the problem is solved, rather than skipping essential reasoning steps. Accuracy never drops by more than 0.5 percentage points, and frequently improves. Ultimately, MoE routing offers both an interpretable lens into reasoning states and a lightweight mechanism for optimizing test-time compute.