Two Decades of Linux Expertise 400-9696-513 24/7 technical support

News

The latest news and hands-on practice from the Linux ecosystem, Xinchuang localisation, cloud native and AI.

China Telecom's Xingchen lab open sources intent model Xing4.0-29B-Intent: one sentence in, structured JSON out

China Telecom AI's Xingchen lab has released and open sourced Xing4.0-29B-Intent, a dedicated intent-understanding model, along with 2.75 million training samples, weights and inference images. It merges intent recognition and parameter extraction, turning colloquial instructions into structured JSON per developer-defined rules and splitting multi-task sentences into separate commands. It already serves the 10000 hotline and smart speakers.

2026-10-22 AI Technology 52 views

Shanghai AI Lab open sources the official AtriaDawn: matched against GPT-5.6 and Claude Opus 5 on eight common benchmarks

Shanghai AI Laboratory has open sourced the official release of its AtriaDawn large model, previously circulated as a preview flagged as unfit for production. Across eight common benchmarks published by the lab it matches Claude Opus 5 and GPT 5.6 sol overall and surpasses them on five, moving into the first tier. Open source, commercial use and private deployment widen base-model options for domestic enterprises.

2026-10-22 AI Technology 46 views

DeepSeek V4.1 Flash upgraded to the 1017 build: wish-style programming lands, in-house search API opens

On Oct 17 DeepSeek updated V4.1 Flash to V4.1-Flash-1017 across web, app and API. Trained with self-feedback reinforcement learning, it can build complete projects from scratch with minimal interaction in game development and 3D modeling, what the community calls wish-style programming. Benchmarks rose while run time and output length fell. An in-house search API at 5 yuan per thousand calls adds cheap web retrieval for agents.

2026-10-20 AI Technology 46 views

StepFun opens Step 5 Preview weights: 600B sparse MoE with a million-token context

On Oct 15 StepFun opened the weights of its Step 5 Preview large model in BF16 and FP8 on Hugging Face and ModelScope, with a technical blog and open source tools. Built for long-horizon agent tasks, it uses a sparse MoE with 600B parameters and 27B activated per token, a 1-million-token context and text-plus-vision input, and ranked second on OpenRouter's weekly usage board before open-sourcing.

2026-10-20 AI Technology 51 views

Arena week 42 large-model blind test: domestic models hold mid-upper positions across fields, kimi-k3-max enters the coding top 10

Arena published its blind-test ranking of AI large models for week 42 of 2026 (Oct 12-18) on Oct 19. The top tier held steady while domestic models kept mid-upper places across boards: kimi-k3-max entered the coding top 10 for the first time and mimo-v2.6-pro the front-end top 20. Three new models joined - gemini-3.5-flash-medium, mimo-v2.6-pro and new flux-2 versions. The rankings rest on anonymous pairwise votes by real users.

2026-10-19 AI Technology 70 views

CAICT releases the top ten AI industry keywords for 2026: AI open source moves toward full stack

At the 2026 AI Industry Conference opened in Jinan on Oct 15, the China Academy of Information and Communications Technology released its top ten AI industry keywords for 2026, spanning technology, application and ecosystem: self-evolving models, AI agent engineering, agentic terminals, embodied AI training, chip-model co-design and open and inclusive sharing. CAICT says AI open source is moving from model weights to full-stack release.

2026-10-19 AI Technology 50 views

Meta open-sources the 816-billion-parameter model Muse Spark Open; two Mac Studios can run it fully locally

Eastmoney reports Meta released Muse Spark Open, an open-weight multimodal MoE large model: 816 billion total parameters, 42 billion active, 1 million-token context window, aimed at coding, multi-step reasoning, and AI agent tasks. Breakthrough: via Apple's MLX it runs fully locally and offline on two top-config M5 Ultra Mac Studios (512GB unified memory) with no cloud GPU cluster, lowering the hardware bar for local deployment of open large models.

2026-10-19 AI Technology 50 views
Previous 1…345 Next 47 items in total
Call us Message
Home Services Consult Message