Search “Multimodal”

CAICT, together with Huawei, vivo, StepFun and others, releases the "Device-Side Intelligent Industry Development Report (2026)"

The Institute of Artificial Intelligence of CAICT, together with the Device-Side Intelligence Group of the Artificial Intelligence Industry Development Alliance of China, and Huawei, ZTE, vivo, Haier, StepFun, ModelBest, Dahua Technology and other organizations, released the "Device-Side Intelligent Industry Development Report (2026)". The report defines device-side intelligence as a fused intelligence form that is primarily local and features device-cloud collaboration, focusing on key technologies such as model lightweighting, heterogeneous hardware integration, foundational software, agents, device-cloud collaboration, and security and trustworthiness. The report also outlines native product forms such as personal intelligence, spatial intelligence, mobile intelligence, industry intelligence nodes and embodied intelligence, as well as application scenarios such as consumer entertainment, smart home, smart office, smart mobility and smart health.

Zhejiang Province Issues the "Guidelines for the Construction of Industrial Agent Platforms (Trial)": Comprising Four Major Modules—Infrastructure, Core Engine, Agent Construction, and Application Interaction

The Zhejiang Provincial Department of Economy and Information Technology has issued the "Guidelines for the Construction of Industrial Agent Platforms in Zhejiang Province (Trial)," specifying that an industrial agent platform consists of four core modules: infrastructure, core engine, agent construction, and application interaction. The platform is built on industrial large models as its foundation, integrates industrial mechanisms and domain knowledge, provides data governance, ontology construction, knowledge learning, model training, and model interaction capabilities, and supports multi-agent collaboration as well as deployment on public clouds, private clouds, hybrid clouds, and edge devices. The guidelines also require the establishment of a guarantee system for security, standards and specifications, and talent cultivation, so as to lower the threshold for small and medium-sized enterprises to apply agents.

iFlytek 1024 Developer Festival Beijing Science and Technology Expo to Open, Spark X2.6 Debuts

The 2026 iFlytek Global 1024 Developer Festival Beijing Science and Technology Expo will open on October 22 at the China National Convention Center Phase II in Beijing, debuting the iFlytek Spark large model X2.6, which completed training and upgrading based on fully domestic computing power, along with on-device lightweight models and a "cloud + device" collaborative solution. The event will also showcase China's first domestic 10,000-card computing power foundation, the Feixing platform, iFlytek Spark Harness, iFlytek AIPC, and AI applications in fields such as education, healthcare, automotive, government and enterprise office, and embodied intelligence.

Fireworks Offers Free Training of Muse Glimmer 30B: Using SFT to Predict Kalshi Market Outcomes

Fireworks AI announced that Muse Glimmer 30B can be trained with SFT for free until November 30, and released a complete recipe for a prediction experiment using historical Kalshi market data. The fine-tuned Glimmer achieved 58.3% accuracy on unseen topics, 11.2 percentage points higher than Jev's 47.1%; in a historical backtest, a hypothetical $1,000 grew to $1,071 after accounting for simulated fees, while both base Glimmer and Jev lost money. Training used LoRA adapters and a K-way cross-entropy loss, ran through the training data only once, and separately used a calibration set to fit a temperature parameter for each question category.

Fireworks Free Weights Club: Training Muse Glimmer to Beat Jev on Predictions

Fireworks AI launches the Free Weights Club, offering free SFT training for Muse Glimmer 30B until November 30. The fine-tuned tuned Glimmer achieves 58.3% prediction accuracy on unseen topics, 11.2 percentage points higher than Jev's 47.1%; in a historical backtest, after deducting simulated trading fees, a hypothetical $1,000 grew to $1,071, while both Jev and base Glimmer posted losses. In a separate synthetic benchmark, Fireworks' System One serving stack delivered median server throughput 2-3x faster than standard BF16 chat completions.

MiniMax Releases M3.1 Flash Model, ALE Score of 29.6% Ranks First Among Domestic Models

MiniMax officially released M3.1 Flash, formerly known as Space Bunny (Yutu), which previously topped the charts anonymously on OpenRouter. The model scores 29.6% on Agents' Last Exam, ranking first among domestic models, and scores 43 on the AA leaderboard; it achieves 73.7% on DeepSWE v1.1, 82.2 points on CyberGym, 76.2 points on BankerToolBench, tying for first with Claude Opus 5, and 61.7 points on OSWorld V2. The model adopts a sparse attention architecture, with output speeds of up to 200+ TPS and a median of 170+ TPS, input at 1 yuan per million tokens, output at 4 yuan, and cache reads at 0.1 yuan, with comprehensive usage costs reduced by 70% compared to the previous generation M3.

SGLang-TorchTPU Open-Sourced: Bringing the SGLang Serving Stack to TPU

The SGLang team, in collaboration with Google and Meta, has open-sourced SGLang-TorchTPU, a PyTorch-based native backend that brings the full SGLang serving stack to Google Cloud TPU and is already in large-scale production use at Meta. Initial support covers models including Muse Spark Open, GLM-5.3, Kimi K3, DeepSeek V4, Qwen3.8, and Gemma 4, with features such as speculative decoding, KV cache offloading, PD disaggregation, and FP8 KV cache. The team also used TorchTitan and Miles to complete GRPO training of Qwen3-4B on a single 4-chip TPU v7x, with the reward curve matching a 6× H200 baseline within random fluctuation.

MiniCPM-V 4.7 1B runs locally on ordinary iPhones

ModelBest's MiniCPM-V 4.7 1B now runs on ordinary iPhones via OnDeviceLLM, no longer limited to the 12GB-memory Pro Max models. The model can perform vision tasks such as reading receipts and calculating totals, extracting flyer information, and understanding screenshots offline on-device, and it also works in airplane mode.

Databricks AI Runtime is now GA: Fine-tune open models on serverless NVIDIA GPUs

Databricks AI Runtime (AIR) is now generally available. Developers can train and fine-tune open models on serverless NVIDIA GPUs without managing GPU infrastructure themselves. AIR supports NVIDIA NeMo, NGC containers, and Nemotron cookbooks, and supports multi-node distributed training, with clusters and networking managed by Databricks. Trained models can be deployed for real-time inference through Databricks Model Serving, and access to training data is governed by Unity Catalog.

English translation is being prepared.

What Is Distributed AI Training? An Analysis of Data, Tensor, Pipeline, and Expert Parallelism

Distributed AI training enables models or batches that exceed the capacity of a single device to be trained by coordinating data, parameters, activations, gradients, and optimizer states across multiple accelerators. Its workflow can be broken down into five steps: partitioning the work according to the chosen parallelism strategy, performing local forward and backward computation, communicating the required tensors, synchronizing or pipelining updates, and checkpointing and recovery across nodes. The article points out that simply adding more devices without changing the communication and memory design is a common misconception; network bottlenecks and synchronization failures can waste most of the theoretical compute.

Franchise Recruitment GEO: An Analysis of the New Customer Acquisition Logic for Franchise Brands in the AI Era

Franchise recruitment marketing is shifting from search bidding and feed ads to GEO (Generative Engine Optimization), with the goal of ensuring that brand facts are accurately mentioned and reasonably recommended in generative Q&A systems such as DeepSeek, Doubao, Tongyi Qianwen, ERNIE Bot, Kimi, and ChatGPT. The article compares the business logic of three models—search advertising, feed advertising, and GEO—and points out that GEO competes on the quality of facts and the credibility of information sources, not on advertising budgets. To judge whether a GEO service provider is reliable, it is recommended to prepare 20 to 50 real franchise recruitment questions and require an on-site demonstration of multi-model queries, brand mentions, cited sources, and erroneous information identification.

English translation is being prepared.

Zhongguancun AI North Latitude Community AI Hackathon opens recruitment: three days of hands-on competition from October 23 to 25

Zhongguancun AI North Latitude Community will hold the "North Latitude Community AI Hackathon" from October 23 to 25, and recruitment is now open. Participants will work in teams to complete topic selection, development, and pitching within three days. The event will award one first prize, one second prize, one third prize, two excellence awards, and one "Best Popularity Award" chosen by audience vote, and will invite AI experts with backgrounds at major internet companies to serve as mentors.

e-Conomy SEA Report: Southeast Asia's Digital Economy to Reach $1 Trillion by 2035

The 11th edition of the e-Conomy SEA report, released by Google, Temasek and Bain & Company, forecasts that Southeast Asia's digital economy GMV will grow 17% year-on-year to $368 billion in 2026 and surpass $1 trillion by 2035. Generative AI is already used daily by 243 million people in Southeast Asia, growing four times faster than the mobile internet, and the report projects that regional AI usage will grow 60-fold over the next five years. 75% of users are willing to let AI agents compare prices and complete transactions on their behalf, but the region will need to expand data center capacity roughly threefold over the next five years to support computing demand.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

Miso Info Tech Participates in 13.8 Billion Won AI Mold Maintenance Project

Miso Information Technology announced on October 20 that it has been selected for the tandem press autonomous manufacturing R&D project led by South Korea's Ministry of Trade, Industry and Energy (MOTIR), with a total budget of 13.8 billion won, led by auto parts manufacturer Sewon Mulsan. Project goals include reducing single mold change time from about 40 minutes to under 5 minutes, achieving a 90% unmanned operation rate and 98% anomaly detection accuracy. Miso will develop a manufacturing AI data platform and predictive maintenance technology based on its IoT Prism platform, integrating equipment, sensor, production and quality data.