Tag
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.
CopT introduces a contrastive on-policy thinking framework for LLMs that generates draft answers first, then uses contrastive verification and dynamic thinking to improve accuracy while reducing token consumption, achieving up to 23% higher accuracy and 57% lower token usage on math, coding, and agentic reasoning tasks.
VideoSeeker introduces a paradigm for instance-level video understanding that integrates agentic reasoning with visual prompts, achieving superior performance through automated data synthesis and reinforcement learning, outperforming GPT-4o and Gemini-2.5-Pro.
ATLAS presents a visual reasoning framework that combines agentic operations and latent representations using functional tokens, enabling efficient training via next-token prediction and reinforcement learning while avoiding intermediate image generation.
This paper proposes an exploration-aware reinforcement learning framework that enables LLM agents to adaptively explore only when uncertainty is high, improving performance on text-based and GUI-based benchmarks.
This paper introduces FlowAgent, a novel framework that reconceptualizes tool chaining as continuous trajectory generation using conditional flow matching to improve robustness in long-horizon agentic reasoning.
Analyzes the 64-D embedding manifold of Google AlphaEarth across 12.1M U.S. samples, shows non-Euclidean structure and poor vector arithmetic, then builds an agentic system with geometry-aware tools that outperforms parametric baselines on environmental queries.