Today for AIIntelligence for practice
Today’s front page8/15/2026

Alibaba Cloud Open-Sources Qwen3.8-27B: Hybrid Attention and Single-GPU Long-Horizon Agents

Alibaba Cloud's Qwen team has open-sourced Qwen3.8-27B, a dense 27B vision-language model. Built on a hybrid architecture combining 48 linear attention layers with 16 gated attention layers, it features a native MTP draft head and controllable thinking mode, delivering native 262K context and single-GPU deployment while advancing SWE-bench Pro and LiveCodeBench benchmarks.

  1. 018/15/2026ExploreTermHybrid AttentionHybrid Attention is an architectural design that combines multiple attention mechanisms across different layers of a Transformer network. It applies linear attention with linear compute complexity to most layers while preserving full gated attention on select layers, balancing inference throughput with reasoning expressiveness across long contexts.
  2. 028/15/2026NewsModelZhipu AI Releases GLM-5.3 with Scaled Post-Training RL and Emergent Cyber Capabilities, Open-Sourcing Weights in Two WeeksZhipu AI has officially released GLM-5.3, a flagship model enhanced through scaled post-training on a 743B base. Powered by IndexShare, SAO reinforcement learning, and the Slime async training framework, the model ranks first across CyberGym (84.5%), AutomationBench (48.2%), and GDPval-AA v2 (1769) while reaching 28.3 on Terminal Bench 3.0, accompanied by a two-week open-source roadmap.
  3. 038/6/2026GuidesIntermediateTool Calling vs. Executing Code: Understanding Code-as-Action in AI AgentsCode-as-Action unifies agent tool invocation and decision-making into executable Python code. This guide systematically analyzes the CodeAct mechanism, execution loop, context shielding, self-debugging, and sandbox defense architecture with leading frontier model selection from SWE-bench Verified and LiveCodeBench, helping engineers build expressive, token-efficient, and secure agent runtimes.
  4. 048/15/2026ExploreModelQwen3.8-27BAn open-weights dense 27B vision-language model open-sourced by Alibaba Cloud's Qwen team. Built on the Qwen3.5 architecture, it features a 64-layer hybrid attention design (48 linear and 16 gated), native MTP draft head, and thinking mode, supporting native 262K context (scalable to 1M tokens) for single-GPU local deployment and long-horizon agentic tasks.
  5. 058/15/2026NewsToolDeepSeek Open-Sources Modular Agent Framework DeepSeek HarnessDeepSeek has open-sourced DeepSeek Harness (dsh), a modular agent framework. Built on the Cordis meta-framework, the system employs a plugin-first architecture to decouple model adapters, execution loops, and tool registries while providing both CLI and local Web interfaces for transparent and extensible agent development.
  6. 068/5/2026GuidesAdvancedExploring Standalone Small Classifiers: Qwen LoRA and BGEThe opening separates the performance problem from the actual base models: locally trainable Qwen3-0.6B and BGE-small-zh-v1.5. For every round from V1 through V4, the article explains the change, measured effect, analysis, and reason for the next iteration, then summarizes final quality and performance; final selection still requires an independent blind test.
  7. 078/15/2026ExploreModelGLM-5.3A 743B-parameter flagship model released by Zhipu AI, enhanced through post-training reinforcement learning scaling with IndexShare, SAO, and Slime for complex terminal execution, coding, and cybersecurity evaluation.
  8. 088/14/2026NewsModelGoogle Releases Gemini 3.7 Flash: Optimized for Long-Horizon Coding and Agentic Workflows with 50% Introductory API DiscountGoogle has officially launched Gemini 3.7 Flash, optimized for long-horizon software engineering and agentic workflows. Featuring a 1M-token context window and tunable thinking modes, the model delivers major gains on benchmarks like DeepSWE and FrontierCode over 3.6 Flash, accompanied by a 50% API price reduction through the end of 2026.
  9. 098/5/2026GuidesIntermediateBuilding a Vector StorageCreating vector storage is the fundamental first step in building high-dimensional semantic search infrastructure. This guide systematically explores document loading, semantic text splitting, embedding generation, and vector database comparative analysis (covering Chroma, Pinecone, Milvus, FAISS, Annoy, and sqlite-vector) to build reliable similarity search infrastructure.