Updates

Google's 8th-gen TPU (training-inference separation) cuts LLM training from months to weeks and boosts inference efficiency by 80%; SJTU's Prof. Yaohui Jin open-sources Path2AGI, a five-dimensional learning map for Chinese AGI education; ex-ByteDance researcher warns widening US-China AI gap amid benchmark-chasing culture masking real-world model usability.

DeepSeek V4 achieves engineering breakthrough with mHC architecture and Muon optimizer—reducing KV cache to 10% of V3.2's at 1M-token context—and fully open-sources code with native domestic chip support. UniWorld-V2.5 matches GPT-Image-2 on dense text and complex layout generation, setting a new benchmark for Chinese AI image synthesis.

The DeepSeek V4 series has officially launched—featuring a 1.6-trillion-parameter Pro version and a 284-billion-parameter Flash version—delivering performance on par with top-tier closed-source models. Notably, it is the first major open model natively optimized for Huawei's Ascend chips, marking a pivotal milestone in China's AI ecosystem's shift away from NVIDIA dependence [11]. Concurrently, the Agent engineering paradigm is accelerating across domains: from intelligent cockpits (by Baizhong, Tencent, and ByteDance) to evaluation frameworks (e.g., Peking University's One-Eval), the 'Model + Harness' approach is supplanting pure model iteration as the core pathway for realizing technical value [12][13][8].

The DeepSeek V4 series has been officially open-sourced, featuring a 1.6-trillion-parameter Pro version and a 284-billion-parameter Flash version—matching top-tier closed-source models in performance and marking the first native support for Huawei's Ascend AI chips, a pivotal milestone in China's push to reduce reliance on NVIDIA hardware [4]. Meanwhile, the release of GPT-5.5 has triggered strategic reinterpretation: OpenAI is explicitly shifting focus toward building an AI 'super-app' ecosystem and strengthening agent-level coding capabilities [20].

In 2026, AI and on-device intelligence enter a new phase—'Agent Post-Training.' GPT-5.5, DeepSeek V4 Flash, and the OpenClaw framework collectively point toward a low-cost, highly deployable path for intelligent agents. Meanwhile, Huawei's Pura 90 Pro Max redefines the entry threshold for imaging flagships at a starting price of ¥6,499, highlighting the accelerating maturity of on-device AI–hardware co-design [1][2][4][5].

Anthropic launched Claude Opus 4.7—centered on 'task resilience' and the ability to respectfully challenge users—while permanently raising rate limits for Pro subscribers, signaling a strategic pivot in large-model competition from 'performance arms races' toward 'trustworthy execution' as the new paradigm.

GPT-5.5 has officially launched—co-designed with NVIDIA—delivering generational leaps in programming proficiency, mathematical reasoning, and agent execution efficiency; it integrates deeply with the Codex platform and GPT Image 2 to build a robust multimodal ecosystem. Meanwhile, Claude's newly launched memory feature and the clarification of the SDK Harness outage reveal that intelligent agent infrastructure is rapidly evolving from 'capable of answering' to 'capable of executing' [8][16][17][3][6].

Chip-level cockpit-and-driving integration and whole-vehicle intelligent-agent operating systems are emerging as new focal points in the intelligent driving race. Horizon Robotics' 'Starry' chip—paired with its KaKaClaw OS—enables natural-language vehicle control, slashing per-vehicle costs by ¥1,500–¥4,000 [3]. Meanwhile, wide-foldable form factors are rapidly reshaping mobile AI interaction paradigms: Huawei's Pura X Max and the rumored Apple foldable iPhone are jointly propelling the industry into a new era of 'large-screen intelligent agents,' centered on content consumption and AI co-piloting [1].

The AI industry is rapidly evolving from isolated tools toward collaborative agent networks. Vast Data's $30 billion IPO valuation underscores surging investor interest in AI infrastructure, while emerging players like Bloome and NeoCognition are pioneering breakthroughs in two key paradigms: Human-Agent hybrid group chat and self-learning general-purpose agents [14][15][13].

OpenAI is accelerating the enterprise deployment of Workspace Agents, while Horizon Robotics has launched the world's first mass-producible 'cockpit-and-driving-integrated' full-stack solution—marking AI's evolution from software-level intelligence to vehicle-wide intelligent agents. Meanwhile, FCGS (Fast Feedforward Compression for 3D Gaussian Splatting) slashes 3D rendering compression time from minutes to seconds, achieving over a 20× compression ratio [10].

OpenAI strengthens its developer ecosystem and engineering capabilities via the Agents SDK and Codex—while rolling out KYC identity verification; Horizon Robotics launches the world's first mass-producible 'cockpit-and-driving-integrated' full-stack solution, ushering in the era of 'vehicle-wide intelligent agents'; Ant Group's Inclusion AI unveils Elephant—a highly efficient large language model with just 100B parameters that tops multiple benchmark evaluations [3][13][12][17].

GPT-Image-2 launches globally with breakthrough Chinese text rendering; WALL-B, the first world-model-based robot foundation model, enables continual learning and autonomy in home environments.

OpenAI fully launches GPT-Image-2—topping LMSYS Image Arena—with stronger complex composition, multilingual text rendering, and real-time data-driven image generation. Google Gemini Deep Research launches two versions with native MCP protocol support for professional data sources.

The convergence of multi-model collaborative workflows and accelerated embodied AI deployment is emerging as a new inflection point in the AI industry: Kimi Claw has pioneered cross-vendor agent group chat collaboration; Variable, Inc. announced its WALL-B robot will enter real households within 35 days—marking AI's shift from 'conversation' to 'presence' [3][4]. Meanwhile, GPT-Image-2's breakthrough performance in Chinese multimodal image-text generation has rendered traditional AI image-detection methods entirely obsolete [2].

Apple's new CEO, John Ternus, is accelerating the deep integration of AI and hardware, while Chinese vendors are driving scalable deployment of AI PCs and multi-device synergy through pre-built AI agents (e.g., YOYO Claw, Xiaomi miclaw) and 'Agent OS' (Kimi K2.6) [1][2][8][21].

The Kimi K2.6 series models have officially launched, igniting developer communities with capabilities including a 300-agent parallel cluster, cinematic-grade frontend generation, and one-click full-stack website creation. Meanwhile, OpenAI Codex introduced its Chronicle feature—equipped with screen awareness and long-term memory expansion—marking a new era of context-adaptive programming intelligence [1].

The AI industry is accelerating into a dual-track era of Agent Native adoption and MoE (Mixture of Experts) practicalization by 2026: Nucleus-Image 17B—the first open-source MoE text-to-image diffusion model—achieves performance surpassing Imagen 4 using only 2B activated parameters; meanwhile, the MCP protocol has been explicitly positioned as the 'connectivity layer' for production-grade Agent deployment, with widespread adoption expected in 2026 [11]. Concurrently, Claude Design is redefining creative productivity boundaries—but its stylistic homogenization and rising token consumption are prompting deep reflection [2][5][3].

Embodied AI achieves a breakthrough validation: Sudo Technology attains a 98% first-attempt grasping success rate under zero real-robot data and zero-shot conditions. Concurrently, the Wish Coding paradigm accelerates adoption—Ant Group's 'Lingguang Ring' democratizes AI-powered app generation for everyday users, marking the practical arrival of intent-based programming [5][10].

AI is shifting toward Agent-Led Growth; brands' discoverability by models like Claude is now a key growth differentiator. MoDA's cross-layer retrieval architecture overcomes Transformer inter-layer bottlenecks, while DeepSeek's >$1B valuation signals China's LLM commercialization inflection point.

Embodied intelligence and AGI infrastructure are accelerating toward real-world deployment: AutoNavi unveiled ABot—the first full-stack embodied AI system designed for AGI—and achieved 15 state-of-the-art (SOTA) results; Moonshot AI, in collaboration with Tsinghua University, proposed the novel Prefill-as-a-Service paradigm, enabling low-latency, cross-data-center KV cache scheduling for the first time [8][7]. Concurrently, industry-wide reflection on the lack of exploratory AI innovation—and criticism of insufficient agent-friendly infrastructure—continues to intensify [10][4].

The launch of Claude Design poses a tangible threat to Adobe and Figma, while the ZJU-REAL team's open-source ClawGUI framework achieves, for the first time, a closed loop of GUI agent training, evaluation, and real-device deployment—marking a new engineering-driven phase in agent deployment [10][2]. Meanwhile, Jensen Huang explicitly refutes the 'de-CUDA' narrative, reaffirming the unassailable moat of the CUDA ecosystem [3].

Embodied intelligence is accelerating toward engineering deployment; FluxVLA Engine emerges as the first open-source, standardized VLA foundation. MiniMax and Alibaba Cloud unveil the 'Model × Harness' paradigm and an enterprise-grade AI Agent security framework, respectively—signaling a pivotal industry shift from isolated capability competition to systemic, collaborative ecosystem building [6][3][11][2].

Embodied intelligence is rapidly entering the 'deployment phase,' with Agibot proposing a new industry stage; dual breakthroughs in world models and VLA (Vision-Language-Action) engineering foundations—Alibaba's HappyOyster and Jizhi Dynamics' FluxVLA Engine—have both gone live; AI Agents are profoundly reshaping frontline software development, and MiniMax's 'Model × Harness' closed-loop ecosystem highlights systemic competitive advantages. [4][19][0][5]

DeepSeek's trillion-parameter V4 model will fully support Huawei's Ascend chips, accelerating the replacement of foreign AI infrastructure with domestic alternatives [5]; meanwhile, the open-source Hermes agent—powered by 'automatic skill evolution'—is rapidly rising to challenge OpenClaw's leadership in the open-source agent space [8]. At the organizational level, AI is reshaping middle management: the emerging concept of 'tack engineering' reveals how AI agent systems are supplanting traditional information relay layers [2].