China AI

AIThe FCC's proposed ban on new Chinese optical transceivers targets the top of the supply stack, but the author argues it leaves the dependencies that matter untouched.

AITrump announced a voluntary agreement under which major AI companies will use internal controls, monitoring, outside audits and board oversight to manage risks.

AIOpenBMB releases MiniCPM-V 4.7 35B-A3B, which it calls the first open-source MoE model to run smoothly on-device without compromising performance.

Why it matters: The source details how Lookahead Expert Scheduling, layer-wise quantization, and speculative sampling aim to run an MoE model on-device, useful context for anyone tracking on-device deployment.

AIKling AI plans to present real-world projects made with its upcoming Kling 4.0 model at ACFM in Busan, Korea, from October 10 to 13, 2026.

AIA data analysis agent built by @Sumanth_077 separates generation, deterministic guardrails, and review: Qwen writes read-only SELECT queries, code enforces hard rules such as a single SELECT, SQLite read-only mode, and a 200-line limit, and a separate TypeSafe AI Jev model checks question clarity, SQL relevance, and whether answers are grounded in returned rows.

AILewis Tunstall says Chinese open models are strong but token-inefficient, citing a plot from the Beam release at IMO. The background post from @reflection_ai says Beam is 3-4x more efficient than GLM 5.2 and over 4x more efficient than leading Western open models in inference.

AIParamount Skydance completed its roughly $110 billion acquisition of Warner Bros. Discovery on October 6, with the combined company renamed Skydance.

AIInferact and the vLLM community reported a 1.9× low-concurrency speedup and about 5.3× throughput under a 150 TPS constraint for DeepSeek-V4.1-Flash over three weeks.

Why it matters: The post breaks down how SWA bounded replay and fused kernels cut prefill and decode costs, a reusable engineering pattern for long-context agentic serving.

AIMiniMax highlighted its longtime creative partner and client SILVERSIDE in a customer spotlight post. The post credits SILVERSIDE with using MiniMax Music and H3 for animation and lip sync work.

AIReflection CEO Misha Laskin argues that closed AI models are like renting an apartment, while open models let users own intelligence as AI adoption grows.

AIClaude Opus 5.5 leads the benchmark with a score of 63.31 at $4.99 per task, ahead of GPT 6 Astra at 56.25 ($11.61) and Sonnet 5.5 at 53.14 ($2.82).

AIMiniMax tagged several partner companies, including Friendli, Anaconda, Kilo Code, Novita, Artificial Analysis, Nous Research, RadixArk, and Vercel, in a post directing readers to its event listings.

AIMiniMax is taking part in SF Tech Week with a series of events on October 6, 7, and 8, featuring partners including Friendli.AI, Anaconda, Kilo Code, Novita AI, Artificial Analysis, Nous Research, RadixArk, Vercel, Fireworks AI, DigitalOcean, Modular, and Evermind.

AIDeepSeek is close to raising at least 80 billion yuan ($12 billion), up from an original target of about 50 billion yuan, according to Bloomberg citing people familiar with the matter.

AIYuchen Jin says Reflection's Beam and Mistral Large 4 both reached roughly GLM-5.2 level within the past two days.

AIStepFun says it is taking part in SF Tech Week, with its presence visible along Highway 101 on the way into San Francisco. The post frames the company as part of the builder, researcher, and team gathering across the Bay Area this week.

AIARC Prize has published full DeepSeek V4.1 Flash results on its public leaderboard, alongside the benchmark code and testing policy needed to reproduce them.

AIDeepSeek says its ARC-AGI-3 evaluations are more operationally intensive than expected, so results will follow once testing is complete. The scores will be reported with a new DeepSeek provider adapter harness.

AIOn ARC-AGI-1, DeepSeek V4.1 Flash scored 88.5% at high reasoning versus 90.5% at low, with high using 35% more output tokens without consistently better answers.

AIDeepSeek V4.1 Flash reaches 72.9% on ARC-AGI-2 at $0.13 per task and 94.5% on ARC-AGI-1 at $0.07 per task, according to ARC Prize verification.

AIKling AI is showcasing videos creators have made with its Kling 4.0 and Kling 4.0 Flash models. The post is promotional and includes no specifications, benchmarks, pricing, or feature details.

AISenseTime says its SenseNova 6.8 Flash Preview model powers Dynamic Design, which turns static images into moving video. The post offers no further technical details, benchmarks, or availability information.

AIMistral says its model outperforms GLM 5.3 in human evaluations across multiple domains. The post gives no benchmark scores, model names, or evaluation details.

AIGeorgi Gerganov says users of Qwen3.8-27B with MTP can get extra speed by switching to DFlash speculative decoding in llama.cpp.

AIOn human evaluation, the model outperforms GLM 5.3 on STEM, CAD, and finance tasks and performs on par on agentic coding. The post is part 5 of a thread, so the model's name and other details come from earlier posts not included here.

AIChina now leads research in nearly 90% of 74 critical technologies, according to the Australian Strategic Policy Institute's December 2025 Critical Technology Tracker.

AIMoonshot AI, maker of Kimi, is planning a Hong Kong IPO in the first quarter of 2027 after closing a private funding round at a valuation of about $50 billion, according to Bloomberg citing people familiar with the matter.

AIHuawei's Mate 90 Pro Max, launched October 1, starts at 9,499 yuan for the 12GB+512GB model, and the Collector's Edition starts at 10,999 yuan.

AIDeveloper Niko1221 open-sourced Strata, an engine that runs quantized Qwen3.8-Flash-Next, a 125B-parameter model, on consumer GPUs with at least 12GB of VRAM. It loads the MoE model into RAM and keeps only frequently used experts in VRAM, and uses a lightweight model for speculative decoding.

AIKhazix on X says he may be seeing things, asking whether Claude now supports Chinese. The post contains no further details, specifications, or official confirmation.

AIMistral has released Mistral Large 4 in Research Public Preview, with open weights for the 1T parameter (49B active) model planned for the end of October.

AIZhipu GLM's GLM-5.3 is now available on Amazon Bedrock, offering enterprises coding and agentic capabilities. The post points readers to the Amazon Bedrock model card for GLM-5.3 for getting-started details.

AIReflection AI, an Nvidia-backed startup, released Beam, its first open-weight large model, aimed at coding and agent tasks.

AINikkei found the average gap between upgraded high-performance model releases among five US and four Chinese developers fell from 125 days (January 2023 to March 2026) to 44 days (April to September 2026).

Why it matters: The piece links faster release cycles to AI systems doing much of their own R&D, a shift that bears on how developers and regulators track model progress.

AIEpoch AI finds that China spends about 2.7 times as much of its final demand on semiconductor producers as the US does, measured as chip revenue per $1,000 of spending in 2022.

Why it matters: The study adds a separated semiconductor input-output dataset, showing China's exposure to chip disruptions is larger than the US's, which helps readers judge supply-chain risk.

AIChinese AI companies earn about 10% of the combined AI-related revenue of OpenAI and Anthropic, according to Epoch AI as of September 2026.

Why it matters: The piece maps how Chinese AI firms earn revenue and why open-weight releases weaken API pricing, giving context for comparing them with US frontier labs.

AIReflection AI previewed Beam, an MoE open model with 501B total and 23B active parameters, claiming 3–4x better inference efficiency than GLM 5.2.

AIRadixArk is speaking at three SF Tech Week events this week on open-source AI, reinforcement learning, and AI infrastructure.

AITencent has open-sourced a new video model called Prism, according to the post. The poster says the previews look solid, with a Hugging Face link provided for access.

AIThe author says Beam, a 501B-parameter open model from Reflection AI, comes close to GLM 5.2 in capability but trails GLM 5.3, Kimi K3, and DeepSeek V4.1 Flash in several areas.