Agentic Reasoning & Long-Horizon Agents

Rapid progress in agentic reasoning with new benchmarks (OmegaUse-OfficeVal, HumanCLAW, ORCA-bench) exposing limitations. Memory innovations (MemHarness, Metis, SkillRise) and practical frameworks (StateAct, Molt, MHGPO) advance the field. Mental World Modeling introduces MENTIS for mental state-aware world models. New: AILMIR agentic multimodal retrieval with incremental learning and metacognitive reflection, showing solid gains on MMMU/ScienceQA. Agent governance and tracing standards remain open challenges.