Free Downloads - Tech News

ConAct framework showing context management as part of the action space for mobile GUI agents

MemGUI-Agent: End-to-End Long-Horizon Mobile GUI Agent with Proactive Context Management

By Hunter Io June 28, 2026AI & Machine Learning

MemGUI-Agent is an end-to-end mobile GUI agent using ConAct framework for proactive context management, achieving 62.5% Pass@3 on MemGUI-Bench with an 8B trained model.

DanceOPD framework overview - composing multiple AI image generation capabilities

AI & Machine Learning

DanceOPD: Composing Multiple AI Image Generation Capabilities Into One Model

By Hunter Io June 27, 2026AI & Machine Learning

DanceOPD is a new on-policy generative field distillation framework that composes T2I, local editing, and global editing into a single flow-matching model using hard-routed field matching.

Agent Memory Systems - Typical Execution Workflows

News

Are We Ready For An Agent-Native Memory System? A Deep Dive Into LLM Agent Memory Architectures

By Hunter Io June 26, 2026News

A comprehensive evaluation of 12 agent memory systems reveals no single architecture dominates all scenarios. Here is what the research found.

Wan-Streamer single Transformer architecture with block-causal attention for audio-visual streaming interaction

AI & Machine Learning

Wan-Streamer: The First End-to-End Real-Time Audio-Visual Foundation Model

By Hunter Io June 26, 2026AI & Machine Learning

Introduction: Beyond Cascaded Systems Current multimodal AI systems are built as cascaded pipelines — separate…

KaLM-Reranker-V1 FBNL framework architecture diagram showing encoder-decoder design

AI & Machine Learning

KaLM-Reranker-V1: A Fast but Not Late Interaction Reranker for Compressed Documents

By Hunter Io June 26, 2026AI & Machine Learning

KaLM-Reranker-V1 introduces Fast But Not Late Interaction (FBNL) for efficient document reranking with Matryoshka Embedding Pooling. Achieves SOTA on BEIR with 10-200x speedup over existing solutions.

FastMix performance comparison chart showing 550x speedup over RegMix

AI & Machine Learning

FastMix: How Tencent Finds Optimal Data Mixture via Gradient Descent

By Hunter Io June 25, 2026AI & Machine Learning

FastMix automates data mixture discovery using gradient descent on a single proxy model, achieving 550x speedup over RegMix for LLM training optimization.

Qwen-AgentWorld: A New Paradigm for Agent Environment Simulation

By Hunter Io June 25, 2026AI

Qwen-AgentWorld is the first language world model simulating agentic environments across 7 domains. Outperforms GPT-5.4 on AgentWorldBench with 58.71 overall score.

AI & Machine Learning

SproutRAG: Attention-Guided Tree Search with Progressive Embeddings — Technical Deep Dive

By Hunter Io June 24, 2026AI & Machine Learning

SproutRAG is an attention-guided hierarchical RAG framework that builds a binary tree over sentence-level chunks using learned inter-sentence attention, achieving 6.1% higher IE with 2.7x fewer tokens than GraphRAG.

AI & Machine Learning Open Source

Sumi-7B: The First Open Diffusion Language Model Trained From Scratch

By Hunter Io June 24, 2026AI & Machine Learning, Open Source

Sumi-7B is the first uniform diffusion language model trained from scratch at 7B scale. This fully open model by Tohoku University can revise any token at any step — a paradigm shift from autoregressive generation.