Latest

MetaCtrl addresses a critical inefficiency in reasoning models by dynamically determining when additional computation is beneficial, eliminating the need for predefined budgets or retraining. This plug-and-play solution significantly improves accuracy while reducing inference costs, making it valuable for developers working with resource-constrained environments. The open-source implementation allows immediate integration into existing reasoning pipelines.

The research challenges the universal compression explanation for post-training effects, revealing instead two distinct regimes affecting mathematical reasoning differently. For AI practitioners and researchers working on mathematical reasoning models, this provides nuanced insights into training approaches and their impact on problem-solving capabilities across difficulty levels and languages.

EASE addresses a critical limitation in agent skill curation by solving cross-executor degradation through behavior adaptation. Developers working with multiple AI models can now use a single, more efficient curator that reduces inference tokens by 9.1-14.5% while maintaining performance across diverse executors, making it particularly valuable for organizations deploying multi-model systems.

This benchmark reveals critical limitations in current AI models' understanding of stateful systems, showing they struggle to track intervention effects without continuous correct feedback. Developers building coding assistants should note that models make compounding errors over time, highlighting the need for robust error handling and state management in AI systems.

ScopeIF addresses a critical gap in LLM training by providing dense supervision for complex constraints that existing binary reward models miss. This approach significantly improves performance on multi-faceted instructions without compromising general capabilities. Developers working with complex task automation can implement this framework to enhance their models' precision, particularly when dealing with segmented constraints across different response parts.

The composition gap reveals a fundamental limitation in how LLMs combine correct intermediate steps into final answers, even when all components are provided. Developers and researchers can use OracleLadder to diagnose specific reasoning failures in their models, with the released code and data enabling reproducible testing across different problem domains.

Training on internalized skill workflows lets a 35B open-weights model beat far larger proprietary models on agentic benchmarks, shifting the edge from raw model scale to curated environment data. Builders running Claude Code or Codex harnesses can reuse the released environments and trajectories for their own SFT or RL pipelines. The results give Qwen-based deployments a concrete benchmark point against Claude Sonnet 4.6 and DeepSeek V4 Pro.

The key finding is that skills can be internalized as model weights rather than injected at inference time, since the fine-tuned agent beats skill-assisted base models even without the skills provided. Agent engineers and fine-tuning practitioners can reuse the released environments and trajectories for SFT or RL with outcome-based rewards. The result also shows a 35B open Qwen-based model outscoring much larger closed models on SkillsBench, a useful cost comparison for teams building coding and tool-use agents.

The core result is that a quantized local model can match a cloud API on Vietnamese compliance extraction, removing the data-sovereignty objection that blocks LLM use under privacy law. Compliance teams and legal-tech builders in Vietnam face a 1 January 2026 deadline, and this shows a single 24 GB GPU suffices for the workload. The unmeasured value-level precision and mid-range token coverage are caveats to weigh before deployment.

Builders running memory-augmented agents on open Qwen3 models can adopt this approach, since it beat strong memory baselines on LoCoMo and PersonaMem without adding retrieval routing or iterative search.

This research provides a new method for improving the exploration capabilities of AI agents, which could lead to more robust and effective systems in various applications.

This research introduces a new AI model for evaluating speech synthesis, which could be useful for developers working on TTS systems.

This provides a new benchmark for evaluating how well large language models can handle complex, multi-step scientific reasoning in interdisciplinary fields like biophysics, which is useful for researchers comparing model performance on specialized tasks.

This research provides a concrete improvement for using AI in financial document processing, showing how to make VLMs more reliable for key-value extraction.

This research provides a new method for aligning large language models with human preferences, which is a key challenge in AI development.

This research presents a new approach to action tokenization that could improve the performance of AI models in robotics and other fields by better preserving the physical relationships between actions.

This research demonstrates a significant improvement in the practicality of running large language models locally by reducing memory requirements.

This research introduces a more efficient way to route skills in large language models, potentially reducing the need for large external parameter counts.

This research provides a method to improve the performance of large language models on GPUs by better managing memory usage, which is relevant for anyone working with or evaluating MoE models.

For those working with camera traps in the field, this study provides crucial data on how well current models can identify species in real-world conditions, showing significant performance drops on low-quality images.

This is a valuable resource for researchers and developers working on multimodal AI, as it provides a new benchmark to evaluate topological reasoning capabilities in large language models.

This paper presents a new parameterization for per-token gating in on-policy knowledge distillation, which is a key technique for training smaller models to mimic larger ones. It provides a framework for comparing different gating designs.