Tag
OpenAI announces the GPT-5.6 model family, including Sol, Terra, and Luna, which achieve frontier intelligence with significantly improved efficiency and cost reductions, backed by innovations in inference and agentic harness.
Robbyant under Ant Group open-sourced the world model LingBot-World 2.0, supporting continuous operation for over an hour with no quality degradation, featuring a brain-cerebellum-like agentic harness, multiplayer support, and multiple interaction actions. Weights and code are open-sourced.
Crucible is an open-source judgment engine that turns a thesis into claims, steelmans them with adversarial tests, measures against a substrate oracle, and iteratively refines the weakest axis, producing verdicts like MATCH, DRIFT, or UNVERIFIABLE.
LingBot-World 2.0 is an advanced world model achieving unbounded interaction horizons, real-time 720p/60fps video streaming, diverse interactive elements, and an agentic harness integrating pilot and director agents. The model is released on Hugging Face with inference code and technical report.
This paper introduces LingBot-World 2.0, an advanced world modeling system with unbounded interaction horizons, real-time 720p 60fps video generation, diverse interactive elements (e.g., attacking, spell-casting), and novel multi-agent behavior control via pilot and director agents, along with a shared multi-player interface.
The VS Code team details recent optimizations to GitHub Copilot's agentic harness, such as prompt caching and tool search, to improve token efficiency and reduce costs under usage-based billing.
Continual Harness is a reset-free, self-improving agentic harness that achieves 20.54% on ARC-AGI-3 at a cost of $774 by storing memories, reusing skills, and refining its prompt, outperforming prior baselines like Hermes and OpenClaw with greater efficiency.
GitHub benchmarked its Copilot agentic harness against model-vendor harnesses, finding comparable task resolution with fewer tokens across multiple benchmarks, highlighting Copilot's support for over 20 models.
Dirge is a Rust-based agentic harness that helps smaller AI models punch above their weight by reducing memory footprint and intelligently managing errors, context, and tool calls, closing the performance gap with frontier models.
An educational deep dive into recursive language models (RLMs), explaining what they are, why they are winning long-context benchmarks, and how they differ from existing agentic harness designs like ReAct or CodeAct, using a simple case study.
HeavySkill is a new framework that internalizes complex reasoning as an intrinsic model skill through parallel reasoning and summarization stages, outperforming traditional orchestration methods and enabling self-evolving LLMs via reinforcement learning.