The paper investigates how users should set the reasoning‑effort level in large language models, a parameter that trades off inference cost for potential performance gains. Experiments on proprietary and open‑weight models across four benchmarks show that higher effort does not consistently improve accuracy, can be up to 42× more expensive on easy tasks, and alters agent behavior by increasing verification and search steps. The authors propose a simple escalation strategy—starting at the lowest effort and only increasing after a verified failure—that matches or outperforms fixed high‑effort settings while reducing cost.
Two prompts can request the same code change and produce the same correct patch, yet cause a coding agent to perform radically different kinds and amounts of work. We study this effect in a preregistered benchmark spanning 4,644 valid runs, 24 deterministic coding tasks, seven reasoning models, and two real agent harnesses.
The paper introduces PORA, a method that teaches large language model agents to decide when to allocate more reasoning steps versus fewer, based on the expected downstream benefit. PORA trains by running each state twice—once with short reasoning and once with long reasoning—and learns to predict the value of extended reasoning. During inference, a controller uses this prediction and remaining budget information to choose the appropriate reasoning length without running both options, improving performance across diverse tasks.
arXiv:2608. 00355v1 Announce Type: cross Abstract: Progress in large language models is often summarized using a single scalar measure, such as a time horizon, a latent ability estimate, or an aggregate benchmark score.
arXiv:2605.09252v3 Announce Type: replace Abstract: Tool-augmented LLM agents tend to call tools indiscriminately, even when the model can answer directly. Each unnecessary call wastes API fees and l...
arXiv:2608. 13571v1 Announce Type: cross Abstract: When a language model fails to answer a query on the first attempt, an agentic system retries, consuming additional tokens each time.
arXiv:2608. 05519v1 Announce Type: new Abstract: Agent benchmarks usually measure task completion and treat resource use as an auxiliary statistic.
arXiv:2608. 20256v1 Announce Type: new Abstract: Reasoning language models trained with reinforcement learning typically operate under a fixed token budget rather than an explicitly adaptive one, which can lead to over-computation on easy problems and insufficient computation on difficult ones.
arXiv:2608. 07885v1 Announce Type: new Abstract: Reasoning modes of language models outperform their non-reasoning counterparts on multi-step agentic tasks, but pay a 3-6x premium in output tokens on every episode -- much of it spent re-deriving procedures that are shared across episodes of the same domain.
The paper introduces the Token Economy Score (TES), a metric that quantifies the accuracy gain of reasoning-capable large language models relative to non-reasoning baselines, normalized by token generation cost. An empirical study across 151 runs on seven diverse benchmarks shows that task structure—such as sequential inference chains—predicts higher TES, while knowledge-recall tasks yield lower TES despite difficulty. The analysis also reveals diminishing returns at higher reasoning effort and highlights how deployment context, via Reasoning Cost Share and Deployment Cost Multiplier, can alter the economic viability of reasoning workloads.
arXiv:2608. 04771v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) excel on complex tasks through long chain-of-thought (CoT) reasoning, but their lengthy intermediate steps cause severe overthinking that inflates inference cost.
The paper introduces agentic meta‑reasoning, a structured inference‑time framework that explicitly manages control decisions—such as selecting partial work, restarting, or stopping—during long‑horizon agentic tasks. By delegating task execution to workers and consolidating decisions through a lightweight controller that references persistent memory, the method reduces the need to replay full histories. Experiments on ProgramBench and other benchmarks show that meta‑reasoning improves performance over direct control baselines, especially as computation budgets increase, and reveals greater reuse of earlier work and higher solution coverage.
The paper introduces the concept of an agent’s "taste"—its ability to make effective long‑horizon decisions—and presents Taste‑Bench, a new benchmark that automatically generates decision‑fork questions from agent trajectories. Taste‑Bench evaluates models on choosing the best path without seeing future outcomes, revealing that top models answer only about 60% of questions correctly and that later‑appearing evidence makes forks harder. The authors also demonstrate that training a student model to mimic a teacher’s judgment improves decision quality and overall success on held‑out software engineering tasks.