SourcesSAGE — Self-Aware Guided Efficient Reasoning Reduces Inference CostarXiv·high signalByteDance. Reasoning models know when to stop thinking but sampling obscures it. SAGE selects concise correct paths via confidence.SourceSource pagearXiv↳ Follow the threadStack layer / Contrast'Resample or Reroute?' Reframes LLM Routing Under Real Budget ConstraintsarXivStack layer / ContrastLong-Horizon-Terminal-Bench: Best Frontier Agent Clears Just 15% of Real Terminal Tasks, Mean 4.3%arXiv (Tencent Hunyuan) / HuggingFace Daily PapersPolicy dependency / Stack layerReason Less, Verify More: Deterministic Gates Catch a Silent Policy-Violation Failure in Tool-Using AgentsarXivStack layer'When Is Routing Meaningful?' Fewer Than 10 Agents Recover Most Diversity in Model SocietiesarXivStack layer'Stop Guessing When to Stop Testing' Pushes Sequential Evaluation to Cut Benchmark CostarXivStack layerzkComposer Decomposes Proof Construction to Scale Zero-Knowledge ML InferencearXivStack layer'Who Broke the System?' AgentLocate Pinpoints the Failing Agent and Step in Multi-Agent RunsarXivStack layer / Threat patternMitigating Taint-Style MCP Server Vulnerabilities Without Touching Server Code (SPELLSMITH)arXiv