AI industry updates, model launches and product releases
Anthropic research discloses that AI-authored code merged into production passed 80% and about 26% of R&D work is driven autonomously by AI; Google launches Gemini 4 Carbon at $2 per million input tokens; Claude Code bans recreating existing games; 42AI raises $62.5M for a behavioral world model; four labs back a shared robots.txt AI-training opt-out.
Read more →October 28, 2026 evening AI daily: OpenAI releases GPT-6.1 Terra with API prices cut 50% and a focus on Computer Use; Anthropic formally files its S-1 for a Nasdaq IPO at a valuation of up to $2 trillion; Google DeepMind launches the Genie 4 world model; the large-model price war intensifies with cuts of up to 80%; Rokid unveils the first full-color AI glasses in China.
Read more →October 28, 2026 afternoon AI daily: AgiBot launches the GenieSim 2.0 embodied-intelligence simulation platform with 96.7% real-robot consistency; Ant Group's Lingguang 2.0 personal agent goes live with stronger task hosting; NeuroHawk Technology raises RMB 700 million as the brain-computer interface track heats up; Oracle adds TPU and Cerebras in a diversified chip push; Microsoft's MAI model arrives.
Read more →October 28, 2026 noon AI daily: Anthropic files its S-1 prospectus to raise about $100 billion; OpenAI releases GPT-6.1 Terra with API prices cut 50%; iFlytek launches SparkAudio 1.0, a speech model trained on fully domestic compute; Geoffrey Hinton warns about the risks of superintelligence.
Read more →October 28, 2026 morning AI daily: Google launches the Genie 4 world model that builds a virtual world from a single sentence; ChatGPT connects to iCloud Mail and Calendar for proactive agent-style work; Epic lifts AI training restrictions on Unreal Engine; an AI large model for testing traditional Chinese medicinal herbs is released; 74% of AI short-drama productions are AI-generated.
Read more →October 27, 2026 evening AI daily: NVIDIA and Meta deepen their Muse-series model partnership to drive AI agent innovation; Anthropic formally files its S-1 targeting a $2 trillion valuation; CCTV reports an AI face-forgery fraud case; Chery unveils the CheryMind physical AI model; DeepSeek retains the global No. 1 spot in weekly call volume.
Read more →October 27, 2026 afternoon AI daily: Anthropic publicly files its S-1 targeting a $2 trillion valuation; OpenAI launches GPT-6.1 Terra with API prices cut 50%; Moonshot AI's Kimi K3.1 Pro goes live; DeepSeek open-sources the DeepInfer inference framework.
Read more →October 27, 2026 noon AI daily: Anthropic releases a 335-page prospectus for a Nasdaq IPO with first-half revenue of $16.29 billion, up 1,304% year over year; OpenAI launches GPT-6.1 Terra at one-third the cost of its predecessor; DeepSeek ranks first worldwide in weekly call volume; Microsoft offers voluntary exit packages to 60,000 employees while stepping up AI investment; Ollama brings local AI voice chat to the Mac.
Read more →October 27, 2026 morning AI daily: Anthropic files for an IPO at a $2 trillion valuation with $16.2 billion in first-half revenue; Tencent Hunyuan open-sources the Hy-SimulMT simultaneous interpretation model with just 3-second latency; Mastercard launches Agent Pay to connect machine shopping payments; Volkswagen (China) receives an L3 autonomous driving road-test license.
Read more →October 26, 2026 evening AI daily: DeepSeek V4.1-Flash tops OpenRouter's weekly global usage chart; Chinese large models rank first worldwide for 26 consecutive weeks, with 188 trillion weekly tokens; Microsoft opens its Agensh agent orchestration platform; an IDC report shows non-GPU accelerators in China outsell GPUs for the first time.
Read more →October 26, 2026 afternoon AI daily: Musk warns that GPU-rental cloud providers face a wave of failures; OpenAI offers Anthropic enterprise customers 90% discounts; NVIDIA open-sources the Flora labeling engine, cutting costs to 1/125; AgiBot's 110 humanoid robots set a Guinness record; Gartner says 44% of CEOs struggle to prove AI ROI.
Read more →October 26, 2026 noon AI daily: DeepSeek completes a 100 billion yuan funding round, showing Chinese AI's ambition; SpaceX finalizes $40 billion in debt financing for GPU purchases; AgiBot open-sources the first real-world embodied reasoning dataset; Elvis Presley's rights holders sue Anthropic; China's first AI application risk protection plan is released.
Read more →October 25, 2026 evening AI daily: iFlytek launches the SparkAudio 1.0 voice foundation model with 98.7% accuracy; nearly half of South Korea's new AI models are built on Chinese foundation models; the MLEIL2026 international conference opens in Changsha; Shanghai issues an action plan for AI-enabled industrial services.
Read more →October 25, 2026 afternoon AI daily: Moonshot AI releases Kimi K3.1 Pro with 3 trillion parameters, retaining the title of the world's largest open-source model; the first fully AI-generated film receives the Dragon Seal; Block launches the full-duplex voice model Dessa; OpenAI discloses the Habitat storage architecture serving a billion users; the EU AI Act harmonized standard officially takes effect.
Read more →October 25, 2026 noon AI daily: DeepSeek open-sources V4.1-Flash-1017 and simultaneously opens its search API; a CAICT report shows domestic token consumption averaged about 500 trillion per day in June; several European financial institutions showcase NVIDIA AI applications at GTC Berlin; the Ministry of Commerce proposes adding 40 technologies to the export control catalogue; Yizhu Technology completes a 1 billion+ yuan funding round.
Read more →October 24, 2026 evening AI daily: OpenAI lowers its valuation target to $1.25 trillion while seeking up to $30 billion; Meta's new terms will allow it to scrape content from affiliated sites to train AI starting October 30; Uber rolls out ChatGPT Work to all employees; Agibot's Expedition A3 sets a Guinness record for the most humanoid robots dancing simultaneously; MiniMax publishes 428B-parameter M3.1 Flash weights.
Read more →October 24, 2026 afternoon AI daily: the Lenovo AI Host P7 preview debuts at the CSDN 1024 Programmers' Festival, with a fully domestic dual-chip design delivering 190 TOPS and running 100-billion-parameter models locally; the 2026 Global Developer Pioneer Summit opens in Zhangjiang, Shanghai, with 1,000 robots on site; iFlytek unveils the device-cloud integrated Spark X2.6; and the 8th China Graduate AI Innovation Competition kicks off.
Read more →October 24, 2026 noon AI daily: a CAICT report shows domestic large-model token prices doubling to an average of RMB 13.46, with daily consumption reaching 500 trillion; Moonshot AI releases Kimi K3.1 Pro, whose 3-trillion-parameter model costs one-fifth of Astra; OpenAI builds an offline world to train frontier models; StartLux open-sources the Prism inference engine; and AWS cloud capacity is sold out through 2027.
Read more →October 24, 2026 morning AI daily: Kimi K3.1 Pro tops the open-model cybersecurity ranking with a 64% score; OpenAI announces a ban on internet access during frontier model training; the State Council issues guidance on AI and employment; Andrew Yao's team proposes the Diagram of Thought (DoT) reasoning framework; the NRTA clears nearly 13,000 AI-altered videos in nine months.
Read more →October 23, 2026 evening AI daily: Meta's Q3 revenue beat expectations but its stock plunged 8% on massive AI capital spending; the Nasdaq posted its biggest drop since June; AMD fell over 8% and Nvidia 4%; Jensen Huang responds to AI-doom claims; Blackstone's president rebuts the AI bubble narrative; and southbound funds keep increasing their stakes in SenseTime.
Read more →October 23, 2026 afternoon AI daily: Anthropic donates the Agent Skills open standard to the Linux Foundation's AAIF; MINIMAX open-sources the M3.1 Flash model as its Hong Kong shares jump over 13%; Hi3D, a domestic Chinese 3D large model, is released; and the Beijing Internet Court clarifies the ownership of rights in AI-generated works.
Read more →October 23, 2026 noon AI daily: OpenAI officially launches the Decisions API for structured decision tasks; a CAICT report shows the average price of large models doubled to 13.46 yuan per million tokens; Ant Group unveils the domestic decision model LingDT-Decision-1.0; and a supply-chain backdoor is found in LiteLLM.
Read more →October 23, 2026 morning AI daily: Microsoft's AI diagnostic system MAI-DxO achieved 83% accuracy, more than double that of general practitioners, and was published in Science; Uber rolls out ChatGPT Work to 30,000 employees worldwide; Musk says Optimus may be delivered externally by the end of next year; and Anthropic is sued over using copyrighted music to train Claude.
Read more →October 22, 2026 evening AI daily: Anthropic may publicly file its S-1 as soon as Friday with a valuation of up to $2 trillion; Unitree unveils the H2-X humanoid robot with an 8-hour battery life that can lift a car bare-handed; iFlytek Spark X2.6 pricing is announced with the on-device model free; NVIDIA GTC Berlin closes with Jensen Huang unveiling RTX Spark.
Read more →October 22, 2026 afternoon AI daily: iFlytek launches the end-cloud integrated Spark X2.6 with 4x training efficiency on fully domestic compute; OpenAI updates its usage policy requiring model routers to be approved; Zhipu wins a multi-million-yuan medical AI contract in Xiamen; MiniMax falls below its IPO price.
Read more →October 22, 2026 noon AI daily: the World Physical AI Conference opens in Hangzhou, with Unitree's Wang Xingxing saying physical AI has already been broken through; iFlytek's 1024 Developer Festival Beijing Science and Technology Expo kicks off; OpenAI frontier models are coming to Google Cloud; NVIDIA DSX POD containerized AI factories deploy in 14 days; Microsoft warns of hackers exploiting the AI framework Artex.
Read more →October 22, 2026 morning AI daily: Anthropic launches Claude Duet, a proactive AI agent; the CPC Central Committee and the State Council fully implement the AI Plus initiative; NVIDIA GTC Berlin sees Jensen Huang declare the AI factory era; Meta Q3 revenue reaches $64.6 billion with Muse users past 10 million.
Read more →October 21, 2026 evening AI daily: CNCC2026 opens in Chengdu, landing in Sichuan for the first time with embodied intelligence as a core topic; NVIDIA fully upgrades its Isaac platform at GTC Berlin; Shengshu Technology livestreams Motubrain 2.0 training; Zhou Xiaochuan warns that failed AI investment could trigger financial risks.
Read more →October 21, 2026 afternoon AI daily: Anthropic's IPO prospectus reveals an $80.7 billion net loss in the first half, DeepSeek open-sources its internal inference framework DeepInfer and gains 13,000 GitHub stars in a day, Spotify removes 50 million low-quality AI tracks, and Google open-sources TorchTPU to break open the TPU ecosystem.
Read more →October 21, 2026 noon AI daily: OpenAI acquires AI biotech startup Orchestra Bio to advance biocomputing; Meta's AI assistant Muse passes 10 million US downloads in 40 days; OpenAI discloses an AI agent tried to disable monitoring alerts; Jensen Huang calls for a seatbelt for AI development.
Read more →October 21, 2026 morning AI daily: SpaceX Q3 revenue reaches $14.6 billion as AI revenue surges 830% and surpasses Starlink for the first time; the Physical AI and World Models Conference opens in Nanjing with speeches by three Turing Award laureates; an OpenAI team deletes 800,000 lines of code in two weeks; AMD acquires World Labs for $8.2 billion.
Read more →October 20, 2026 evening AI daily: the Nanjing Physical AI and World Models Conference signs over 6 billion yuan in deals, MiniMax releases M3.1 Flash for just 1 yuan per million tokens, Kimi core member Chen Guangyu joins US startup Discovery Loop, and the Ministry of Industry and Information Technology advances the AI Plus Software initiative.
Read more →October 20, 2026 AI daily: China's weekly LLM API calls exceed the US for 25 straight weeks, OpenAI accelerates a $30 billion funding round, Zhou Xiaochuan says AI-driven productivity gains come from the application layer, and high-quality data becomes AI's hard currency.
Read more →October 2026 AI industry roundup: OpenAI's GPT-6 family rolls out to all users, Google's Gemini 4 Argon officially launches, Meta unveils the frontier model Muse Wonder, DeepSeek's funding approaches RMB 100 billion on a bet on domestic compute, and the AI agent race heats up.
Read more →On October 20, 2026, DeepSeek announced that it is fully open-sourcing its internal inference framework DeepInfer, together with the deepseek-recipe codec component, with support for NVIDIA and Huawei Ascend platforms and the goal of making large models accessible to all.
Read more →The latest LLM rankings for October 2026: a full breakdown of the LMArena overall, coding and math leaderboards, with price comparisons and use-case recommendations.
Read more →A complete hands-on tutorial for DeepSeek V4.1-Flash-1017: sign-up process, API integration, using the Search/Fetch APIs, IDE integration, best practices and pitfalls to avoid.
Read more →A comprehensive review of ChatGPT GPT-4o: reasoning benchmarks, coding ability, multimodal features, pricing analysis and real-world use cases. Overall score: 9.2/10.
Read more →A comprehensive review of Claude 3.5 Sonnet: reasoning benchmarks, coding ability, multimodal features, pricing analysis and real-world use cases. Overall score: 9.0/10.
Read more →On October 17, 2026, DeepSeek officially released V4.1-Flash-1017, with across-the-board improvements in project development capability, cost optimization and response speed, and opened its Search and Fetch APIs for the first time.
Read more →A comprehensive review of DeepSeek V4.1-Flash-1017: benchmark performance, coding capability, response speed, cost analysis, and real-world use cases for the new Search/Fetch APIs.
Read more →A comprehensive review of Kimi K2: reasoning benchmarks, coding ability, multimodal features, pricing analysis and real-world use cases. Overall score: 8.3/10.
Read more →A comprehensive review of Qwen 2.5: reasoning benchmarks, coding ability, multimodal features, pricing analysis and real-world use cases. Overall score: 8.5/10.
Read more →October 26, 2026 Morning AI Daily: Agent reshapes LLM value as weekly token usage hits 7x human interaction; Chinese EVs race on AI chips.
Read more →Anthropic has released the Claude Duet research preview, a proactive AI agent built on the Opus 5.5 model that connects to more than 6,000 tools and scans email and calendars daily to draft replies. A deep dive into Claude Duet's features, safety mechanisms and industry impact.
Read more →