Latest

This model is competitive with frontier agents and achieves the highest reported score on five of 16 benchmarks spanning real-world research, engineering, and digital work.

This research presents a new multimodal approach for slip detection in robotics, achieving high accuracy and low latency, which is relevant for improving the performance of robotic systems.

For users interested in optimizing large language model inference on AMD hardware, this paper provides detailed engineering insights and performance benchmarks.

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.

This research presents a new approach to improving code generation efficiency, which is relevant for developers and researchers working with large language models.

This research presents a new approach for robot action prediction that could improve the performance and efficiency of robotic systems.

This is a new optimizer that could improve training efficiency for AI models.

This research introduces a new approach for dynamic optimization problems, which is relevant for anyone working in fields like logistics, resource management, or scheduling where conditions are constantly changing.

Show-Harness enables VLMs to control robots with minimal fine-tuning, making advanced robotics accessible without specialized hardware or extensive training.

The 10% full-task completion rate is 3.3 times higher than previous agents on this benchmark, showing significant progress in autonomous research agents.

Gander's open-source release with models, code, and data provides developers with a framework for building real-time AI agents that handle complex workflows.

PlannerForge improves autonomous driving testing efficiency, with open-source models matching commercial performance and reducing collisions by over 50% through cost-tuning.

SkillAdam reduces the cost of developing agent skills by requiring fewer optimization iterations, making it more efficient for developers building AI agents.

RedKnot-MLA improves DeepSeek-V4's long-context efficiency with 2-4x faster response times and significant computational savings for large document processing.

The 9B model reduces token usage by 60% in hybrid tasks, offering developers a more efficient approach to computer automation than GUI-only methods.

ACE improves efficiency for MoE models without retraining, benefiting developers working with large language models in resource-constrained environments.

Enterprises can deploy Sentinel-RL to handle large-scale security operations with faster processing and higher accuracy than traditional methods.

Developers can use Terminal-Universe to create more realistic training environments for terminal-based code agents, improving their performance on real-world coding tasks.

The framework outperforms direct LLM reformulation in supply chain operations, offering retailers a more reliable solution for adapting to evolving requirements.

Jina-OCR-v1 offers efficient document parsing accessible to budget-constrained users in ASEAN with its open-source availability and performance on low-cost hardware.

This approach improves AI research efficiency by converting open-source code into reusable skills that boost performance across multiple benchmarks.

HoH achieves 52-83% performance gains over existing coding agents, potentially accelerating software development across ASEAN's growing tech sector.

H3-World enables precise video control through language instructions with minimal training data, making advanced video generation accessible for developers in the region.

Facet-0 enables precise robotic assembly at sub-millimeter tolerances, outperforming existing baselines by over fivefold in success rates.

Released under Apache-2.0 license with lowest median latency at 14.3 ms, offering robust protection against Russian and English prompt injections.

The method enables academic teams to perform AI-powered drug discovery on consumer hardware, previously requiring expensive GPU clusters.

CineForge reduces computational costs by 37% while improving long-form video generation quality, making it valuable for content creators in resource-constrained markets.

Researchers can use ASKS to map research trajectories and track knowledge evolution across papers, with all changes traceable to original sources.

This open-source training recipe enables affordable large language model development for researchers and developers worldwide.