Jul 2026· Journal of Science Innovation and Technology Research· 0 citations
TL;DR
Fresh, objectively scored benchmark items can support auditable accuracy-cost-latency routing when features encode verifiable computational structure, and show that fresh, objectively scored benchmark items can support auditable accuracy-cost-latency routing when features encode verifiable computational structure.
Abstract
Dynamic benchmarks make model selection harder and more useful at the same time: a router must identify when a low-cost policy is reliable without learning benchmark-specific artifacts. This study evaluates prompt-level routing on 200 LiveBench reasoning questions (100 zebra puzzles, 50 spatial questions, and 50 Web-of-Lies questions). Two reproducible, deterministic reasoning back ends are compared. FastLite uses exact reasoning only for recognized low-complexity cases and bounded heuristics otherwise; SolvePro uses full constraint, Boolean, and analytic spatial procedures. A router predicts whether FastLite is safe from task, length, release, constraint, and structural features available before an answer is generated. SolvePro reaches 100.0% exact accuracy, whereas FastLite reaches 67.0% exact and 69.9% answer-component accuracy. Under five-fold out-of-fold evaluation, a logistic router reaches 97.0% exact accuracy with a 32.5% strong-call rate and a 55.0% normalized cost saving relative to always using SolvePro. A transparent complexity rule preserves 100.0% exact accuracy with a 36.5% strong-call rate and 52.7% cost saving. Errors are concentrated in spatial and zebra items, confirming that prompt length alone is insufficient. The findings show that fresh, objectively scored benchmark items can support auditable accuracy-cost-latency routing when features encode verifiable computational structure. Because both back ends are deterministic policies rather than hosted LLM APIs, the reported latency and normalized cost values are controlled experimental quantities, not vendor performance claims.
Test-time scaling is often implemented by spending more compute along one axis: sampling more solutions, extending a chain of thought, or applying a stronger evaluator. Under a fixed inference budget, these choices compete. This paper formulates test-time reasoning as a compute-allocation problem in which a system must decide whether the next unit of compute should be spent on generation, verification, or stopping. We introduce CoBa, a compute-balanced routing policy that first obtains a small set of candidates, applies cheap verification broadly, and routes uncertain or high-value candidates to stronger verification. On 3,129 example-generator evaluations spanning MATH-500, AIME 2024/2025, AMC 2023, and procedural symbolic reasoning, CoBa-Routed-Strong reaches 85.13% macro accuracy, statistically matching a self-evaluation weighted-voting proxy at 85.20% while using 49.1% fewer parameter-weighted tokens. It also matches best-of-16 majority voting within 0.01 macro-accuracy points while using 58.9% fewer parameter-weighted tokens; paired tests retain a small best-of-16 edge at substantially higher cost. Paired bootstrap tests show significant gains over single-sample decoding, while the remaining gap to the pool oracle exposes headroom for sharper routing. For local reasoning systems, test-time scaling becomes a question of where the next computation is most valuable.
Yan Zhou, O. Yue, Kaiyang Zheng et al.· 0 citations
CRM+RCCR, an architecture-agnostic cost-aware objective that encodes cost preference into continuous relevance targets through per-pair independent scoring, eliminating multi-positive dilution while regularizing queries with similar routing preferences to be closer in the routing space.
This work introduces Test-Time Scaling via Error Localization (TTEL), an inference-time algorithm that utilizes fixed or environment feedback to perform token-level error localization and establishes strictly dominating Pareto frontiers across sequential reasoning domains.
Static load balancers cannot mitigate a backend that is degraded rather than down: round-robin and least-connections keep routing traffic to a server returning HTTP 500s until an operator intervenes. We ask whether a Large Language Model can replace the static routing policy itself, reading HAProxy and Prometheus telemetry every 10 seconds and isolating faulty servers through guardrailed calls to the HAProxy Data Plane API. On a reproducible benchmark with a persistent structural fault built into roughly one-third of a heterogeneous fleet, we sweep 15 open-weight models across five families (0.35B to 35B total parameters; dense, mixture-of-experts, and efficient-sparse architectures), reasoning modes, fleet scales of 3 to 9 backends, and two routing algorithms, totaling 240 runs. We find a capability threshold near 3B active parameters. Below it, LLM policies are typically unreliable and sometimes worse than no policy; above it, every model, regardless of architecture, saturates near an 88% reduction in client-perceived 5xx errors over the static baseline. The threshold is approximate: Gemma 4 E2B clears it with 2B active parameters, while the dense 3B Granite 4.0 Micro does not. The availability gain has costs. Draining concentrates load onto surviving servers, inflating tail latency 2.6 to 2.8 times, and enabling reasoning multiplies token spend roughly tenfold, overrunning the control interval and degrading effectiveness. The efficient operating point is a supra-threshold model in its cheapest non-reasoning mode, wrapped inside deterministic guardrails.
Benchmark gains are often treated as evidence of greater LLM capability. Yet the same gain can reflect different changes in model behavior. A model may reach new answers, or produce answers that were already within reach. Aggregate scores do not distinguish these changes question by question. We establish a question-level audit under fixed budgets, temperatures, and answer formats. A question is realized when the default deployment procedure produces the correct answer. A question is reachable when a specified probe finds that answer within a fixed budget. We first test whether inference-time layer routing can expand reachability. Under a matched budget, random routes match or exceed structured search in all 43 model and task settings. Answer-blind procedures retain almost none of this gain, which instead requires access to the correct answer. We then ask why reachable answers sometimes fail to appear. Across six cases spanning 0.5B to 31B, silencing one identified MLP block repairs 68 to 92 percent of a predefined failure set. We next test whether training closes the gap by expanding reachability. In five of six matched evaluations, deployed performance rises while the reachable ceiling remains flat or falls. For DAPO, the deployed score rises by 14.7 points while the reachable ceiling falls by 13.3 points. Across the settings we audit, realization and reachability therefore do not always change together. Claims of capability expansion should report both realized performance and reachability under matched evaluation conditions. Code is available at https://github.com/LiZaiyuan0619/reachability-not-realization
Yanchao Li, Wanhao Liu, Jiaqing Xie et al.· 0 citations
LLM-Advisor is introduced, an open-source, adaptive framework designed for intelligent query categorization, dynamic model evaluation, and constraint-aware request routing across heterogeneous multi-LLM pools that achieves a 42% reduction in overall inference expenditure and a 35% decrease in average response latency.
Harshil Lodhiya· International Journal of Res...· 0 citations