agentic-reasoning

DeAR is a decentralized agentic reasoning framework that enhances accuracy in knowledge-intensive reasoning tasks through capability grounding and collaborative thought navigation, outperforming centralized methods on multimodal benchmarks.

This paper presents a unified global-to-local paradigm for video anomaly detection, introducing a training-free framework (GtS) and a tool-augmented agentic reasoning method with reinforcement learning, along with a new benchmark VAGU-T and metric JeAUG.

QuantumMind presents an auditable agentic workflow that automatically generates and screens quantum speedup hypotheses using typed role-specialized actions and a deterministic validator.

Maxime Labonne highlights early fine-tunes of LFM2.5-2.6B, while Bad Theory Labs releases two open-weight models: BTL-4 35B, a frontier agentic reasoning model, and Macaw 2.7B, an on-device Mac agent, with notable BFCL v4 performance.

CastFSR is a Fast–Slow–Reflect agentic reasoning framework that leverages LLMs for context-aware time series forecasting, combining fast lightweight forecasters, slow deliberative reasoning, and reflective evaluation to improve forecasting accuracy and consistency.

Introduces agentic graph token reasoning, a method that recasts graph tokenization as part of an LLM's step-by-step reasoning process, allowing the model to dynamically choose graph views and granularity during inference. The method outperforms baselines across seven graph domains and transfers zero-shot to unseen domains.

This paper introduces LedgerMind, a provenance-constrained multimodal agentic reasoning framework that uses a Structured Evidence Ledger to ensure grounded, faithful reasoning in visual question answering, addressing failure patterns like hallucination and over-reasoning.

SCAIR introduces a training-free framework for reasoning over enterprise knowledge graphs by injecting schema-conditioned structural priors and enforcing schema-aware traversal during multi-hop reasoning, showing improvements on a real-world CMDB benchmark.

Legora introduces the Legora BAR benchmark for evaluating AI agentic reasoning in legal workflows, using real cases and the Legora harness to measure system-level performance.

This paper introduces FinanceComplexQA, a comprehensive benchmark for evaluating agentic reasoning on industrial-grade financial documents, featuring bilingual support, expert-level questions, and complex layouts across six scenarios and seven tasks.

This paper surveys the emerging paradigm of Geospatial Foundation Models (GeoFMs), which are pre-trained on massive geospatial datasets to enable rapid fine-tuning and zero-shot analysis of satellite and aerial imagery. It covers the paradigm shift, model adaptation strategies, and a forward-looking vision of Agentic Geospatial Reasoning using LLMs as orchestrators.

AGORA is a new benchmark for evaluating large language models on archive-grounded reasoning tasks across workplace documents, comprising 362 questions over 9,664 real documents. The strongest model achieves only 59.4% accuracy, highlighting substantial room for improvement.

Nemotron 3 Ultra is a 550B parameter hybrid Mamba-Attention mixture-of-experts language model, pre-trained on 20T tokens, extended to 1M context, and post-trained with SFT, RL, and MOPD. It achieves up to 6x higher inference throughput than state-of-the-art LLMs with comparable accuracy, and is open-sourced.

PhotoCraft proposes a training-free hierarchical memory system for photo-search agents, integrating working, episodic, and semantic memory to maintain long-horizon context and transfer knowledge across tasks, achieving up to 18.5% improvement on DISBench.

This paper introduces Adaptive Latent Agentic Reasoning (ALAR), a dual-mode framework for LLM agents that uses compact latent reasoning for routine turns and selectively escalates to explicit chain-of-thought for harder decisions, achieving up to 84.6% token reduction while maintaining task accuracy.

This paper introduces DAR (Deontic Agentic Reasoning), an agentic framework enabling LLMs to interactively query statutes and policies for legal/regulatory reasoning tasks. Evaluated on DeonticBench, results show agentic harnesses improve frontier models but can degrade weaker models on numerical tasks while consuming more tokens.

CP-Agent presents a calibrated risk-controlled approach for feedback-driven competitive programming using large language models, achieving significant improvements on benchmarks without parameter updates.

This paper proposes SAM, a state-adaptive memory framework that dynamically manages interaction histories for long-horizon agentic reasoning, enabling intent-driven recall without retraining the backbone model. It outperforms strong baselines across multiple benchmarks like BrowseComp and HLE.

OpenAI's GPT-5.5 model shows significant improvements in complex agentic tasks and code generation, outperforming previous versions and competing models like Claude Opus.

Introduces SR²AM, a framework for efficient agentic reasoning via self-regulated simulative planning, achieving competitive performance with models 20-30x larger while using 26-95% fewer reasoning tokens.