China AI

AIPurple Mountain Laboratories, China Mobile, Jiangsu Future Network Group and the Jiangsu Future Network Innovation Research Institute released AXP, an agent interconnection platform built on deterministic networks, at the 10th Future Network Development Conference in Nanjing on October 16.

AIDeepSeek released V4.1-Flash-1017, an updated open-weight Flash model that DeepSeek says improves on its predecessor in every tested category.

AIDeepSeek released a new post-training checkpoint of its V4.1-Flash model, V4.1-Flash-1017, available through the deepseek-flash name on its website, app, and API. DeepSeek reports 77.0 on DeepSWE v1.1, above Claude Opus 5's 74.0 in its own table, and says the checkpoint beats the previous V4.1-Flash on all 11 benchmarks it reported.

AIDeepSeek released DeepSeek-V4.1-Flash-1017, which the company says outperforms its predecessor on benchmarks, finishes tasks faster, and produces shorter outputs.

AIvLLM congratulates DeepSeek on DeepSeek-V4.1-Flash-1017, which the post says has better scores and shorter outputs for long agent loops. vLLM says it delivers 1.3B tokens per dollar TCO on SemiAnalysis AgentX, at 830K tok/s per GPU (cached + input + output) at 100 tok/s per user.

AIDeepSeek released DeepSeek-V4.1-Flash-1017, which it describes as a Pareto improvement with stronger benchmarks, faster task completion, and shorter outputs.

AIDeepSeek says DeepSeek-V4.1-Flash-1017 outperforms its previous version on benchmarks, completes tasks faster, and produces shorter outputs.

AIDeepSeek has released DeepSeek-V4.1-Flash-1017 with open weights on Hugging Face. The post gives no benchmark scores, parameter counts, or context length.

AIDeepSeek says its DeepSeek-V4.1-Flash-1017 model outperforms the previous V4.1-Flash on benchmarks, completes tasks faster, and produces shorter outputs.

AIMeng Shao recommends jobs.dinq.me, which lists AI jobs from companies with over $10M in funding, listed internet firms, quant firms and neo labs.

AIDeepSeek has released V4.1 Flash 1017, which Guizang says significantly improves overall capability while shortening task run time and output length at lower cost.

AIDeepSeek released DeepSeek-V4.1-Flash-1017, an open-weight model that scores 77% on the DeepSWE 1.1 benchmark, according to X: Testing Catalog.

AIVercel says DeepSeek V4.1 Flash 1017 is now available on AI Gateway, with gains in reasoning, coding, agentic tasks, and images. The model is also said to finish tasks faster and use tokens more efficiently.

AIDeepSeek says DeepSeek-V4.1-Flash-1017 delivers stronger benchmark performance, faster task completion, and shorter outputs than the previous version.

AIThe China Meteorological Administration has put the AI-based global ensemble forecast system CMA-AIM-GFS-Fengqing-ENS-V1.0 into national operational use after an operational review.

AIOpenRouter has added DeepSeek V4.1 Flash 1017, with DeepSeek as the first provider and more providers planned. DeepSeek says the update is more token efficient and shows major gains on agentic benchmarks.

AIDeepSeek is opening the search engine behind its own app to developers through new Search and Fetch API endpoints.

AIMoonshot AI has teamed up with Agricultural Bank of China and American Express on what it calls "the world's first AI native credit card," which grants Kimi membership benefits.

AIDeepSeek says V4.1 Flash 1017 is its October 17, 2026 snapshot of DeepSeek V4.1 Flash, served under the deepseek-flash API model name. V4.1 Flash is a sparse mixture-of-experts model.

AIStepFun has released Step 5 Preview's FP8 and BF16 weights on Hugging Face and ModelScope, a sparse MoE model with 600B total and 27B active parameters per token.

AIDeepSeek has upgraded its V4.1 Flash model to DeepSeek-V4.1-Flash-1017, with web and app models updated too.

Why it matters: The update lists benchmark gains alongside lower task runtime and output length, showing how the model's cost-performance balance changed from its predecessor.

AIDeepSeek says its in-house search engine, which handles hundreds of millions of requests daily, is now available through its API. The engine indexes hundreds of billions of web pages, supports multilingual search, and filters SEO spam, GEO manipulation and low-quality AI-generated content.

AIDeepSeek releases DeepSeek-V4.1-Flash-1017 on Hugging Face as a minor update to DeepSeek-V4.1-Flash.

Why it matters: The source gives benchmark scores against several named rival models, showing where this release sits on performance and cost in agent tasks.

AIMagnific posts a Kling 4.0 video showing flying cars picking people up. The post itself gives no further technical details, pricing, or availability.

AIDeepSeek published Harness v0.2.1-rc1, the first release candidate of the 0.2.1 series for its open-source agent runtime.

AIShanghai Artificial Intelligence Laboratory has released the full version of Atria Dawn, an agent-oriented model built on the 744-billion-parameter GLM-5.2 base, with a context window raised from 256K to 512K tokens.

AIPurple Mountain Laboratories and China Mobile have launched AXP, an agent interconnection platform built on deterministic networking.

AIindigo (@indigox) says Muse manages personal life and Grok Bot manages work, arguing Chinese walls are too high for personal agents to cross.

AIDesign Arena projects Reflection AI's Beam 1.0 to rank #63 on its Website leaderboard with an Elo of 1240.

AIKling AI's X post asks who needs a kitchen torch and tags the Kling4 and KlingAI hashtags, promoting the Kling 4 model. A reply from @magnific shows a video of a kitchen torch caramelizing a dessert, with the caption "Wait for the crack."

AIKling AI promotes its upcoming Kling 4.0 video model with the hashtags #Kling4 and #KlingAI and a short post describing a scene as somewhere between fantasy and nostalgia.

AIShenchen Kaidong's badminton robot completed 1,452 consecutive rallies with a human player on December 22, 2025, earning a Guinness World Record for most consecutive counter hits by a mobile robot.

AIBaidu's official account has outlined five highlights for Baidu World 2026, to be held November 12 at the Beijing National Convention Center Phase II under the theme "The Next Act, Intelligent Organizations." The event will include one main forum, 12 breakout forums, 30+ AI open classes and a 6,000 m² AI exhibition area, and Robin Li will deliver the keynote.

AIUBTech says its Thinker-WM ranks first on WorldOlympiad with a 0.740 composite score, and leads EWMBench with 4.66 and WorldArena Track-1 across 12 metrics.

AIAnt Lingbo's pharmacy robots, deployed in Shanghai, use its general brain, which was pretrained across 17 robot brands and 20 configurations, to pick medicine boxes.

AIWu Haixu, first author of the GeoPT paper, has founded Wuyuan Technology, a world-model startup that has raised several rounds and reached a valuation of several hundred million US dollars, with investors including Ant Group and Meituan.

AIDeepSeek has added Search and Fetch endpoints to its API platform, designed to give agent and RAG applications live web access without a third-party search provider.

AITencent Hunyuan has released Hy Memory 2.0, open-sourcing its memory model, framework and local app MemHub under Apache 2.0.

AIUBTECH says its Thinker-WM world model encodes RGB video, metric depth and robot actions in one shared representation, and that it ranks first on three embodied world model benchmarks in its own evaluation.

AIUnitree Robotics has released UniUni, a free desktop swarm control app for Windows and macOS that supports its H2 and G1 humanoid robots.