VHD-Play is a pipeline that generates diverse agentic reinforcement learning environments by sampling and solving mathematical models before rendering their decision processes as stateful tools. This approach ensures that executable dynamics and trajectory-scoring references are inherited directly from the solved model, addressing the misalignment issues common in existing generation pipelines.
The authors introduce WebMRE, an offline benchmark of 541 tasks and 5,293 steps derived from successful WebArena trajectories, designed to provide a deterministic protocol for scoring web agent checkpoints without environment drift. This framework pairs human-oriented guide sentences with grounded actions to study the mutual reinforcement effect between them.
Researchers introduce SkillGym, a framework that transforms human-written agent skills into executable, verifiable training environments for large language model agents. The system utilizes a skill-to-task pipeline to instantiate concrete tasks and verify outcomes with code-based checkers.
Airbnb has entered a new agreement to provide its engineering and product development teams with broader access to OpenAI frontier models, including GPT-6 Astra. This expansion builds on the company's existing use of Codex and collaboration with OpenAI, extending access through OpenAI APIs and Amazon Bedrock.
The Goose desktop application has released version 1.52.0, introducing live voice conversations as a primary new feature alongside support for several updated AI models.
Claude Code version 2.1.281 introduces new capabilities for the Claude apps gateway, including support for newer Claude Desktop keys in policy blocks, IAM role assumption on Bedrock upstreams, and configurable guardrails. The update also enhances the CLI with MCP URL-mode elicitation for browser-based flows and validation checks for plugin configurations.
invideo utilizes GPT-6 Astra to enhance its agentic video editing capabilities, achieving a threefold increase in color grading success rates and producing 50 custom effects in a single day.
Anthropic has announced a new life sciences research group dedicated to using Claude to identify uncharacterized protein families and test hypotheses through laboratory experiments. In early results, the AI autonomously discovered a novel enzyme system called array-associated reverse transcriptases (ART) found in bacteriophages.
Google is updating its Private AI Compute platform to support persistent, cross-device AI memory while maintaining on-device privacy standards. This change resolves the dilemma of providing long-term continuity for AI assistants without compromising the strict privacy limits typically associated with on-device processing.
A developer has released an open-source experimental framework called Gear to explore recursive self-improvement (RSI) and is seeking community feedback on its design.
Anthropic has released Claude Projects v2, redesigning the feature from a folder-based structure to a conversation-driven interface that manages builds through task delegation and parallel threads. Simultaneously, Google is testing a family agent that operates on a dedicated cloud computer to coordinate activities for up to six household members.
This update covers several developments in the AI landscape, including Meta opening its Muse platform to third-party developers, releasing a new version of its segmentation model, and a security incident involving Google's Gemini.
Xiaomi has open-sourced the MiMo-V2.6 Pro and Flash omnimodal models, which are designed to coordinate agents for constructing and visually testing interactive 3D scenes. These models can generate Blender assets, control robotic arms from camera feeds, produce frontends and presentations, assemble videos, and compose music as scores and MIDI.
Researchers present AIDE^2, a system that implements a loop of recursive self-improvement for an AI research agent by optimizing its own code. The system proposes changes to its internal logic, benchmarks modified versions on AI R&D tasks, and retains only those improvements that perform best on hidden evaluations.
The authors introduce Growing Harness, a training paradigm that learns an agent's control structure from task feedback rather than relying on standard context-heavy harnesses. By converting recurring control decisions into executable code and reserving LLM calls for semantic reasoning, the method aims to create reusable specialist agents.
Researchers introduce CliffCompaction, an autocompaction technique designed to reduce costs for long-horizon coding agents by up to 50% under a bounded context. The method maintains or improves performance on Terminal-Bench and achieves state-of-the-art results on KernelBench by keeping compacted information faithful through truncation rather than rephrasing.
Researchers present AIDE^2, a system that implements a loop of recursive self-improvement for a frontier AI research agent by optimizing its own code. The system proposes changes to its internal logic, benchmarks modified versions on AI R&D tasks, and retains only those updates that perform best on hidden evaluations.
Researchers introduce CliffCompaction, an autocompaction technique designed for agents handling millions of tokens, which cuts costs by up to 50% under bounded context. The method maintains or improves performance on Terminal-Bench and achieves state-of-the-art results on KernelBench by keeping information faithful through truncation rather than rephrasing.
Researchers introduce Agensh, a scalable self-organized multi-agent harness that eliminates the central orchestrator bottleneck by allowing concurrent workers to asynchronously claim sub-tasks and merge progress.
Alibaba Cloud has introduced Qwen3.8-Omni-Flash, a natively multimodal agentic model designed for real-world productivity tasks that substantially improves multimodal understanding and reasoning compared to previous omni models.