This issue collects 44 AI updates from 2026-10-20, led by Developers are using AI agents to strip ads and bloatware from smart TVs, Tencent's Hy Memory Claims Open Source Benchmark Lead Over Claude Opus 5, Tolan raises $30M for its AI companion as active users pass 250,000, Replika's Kuyda opens Wabi to all users: a multi threaded, shareable AI assistant.
Developers and vibe coders are pointing agents like Claude Code at their televisions to remove system ads, disable viewing tracking and rewrite the launcher, with public GitHub projects such as dumbtv and cleanyourtv leading the charge.The work needs no root access, and thousands of owners have already tried it on sets from Xiaomi, TCL, Sony, LG and Fire TV.
Replika founder Eugenia Kuyda has opened Wabi to all users: a personal AI assistant built as a messaging app, with multiple shareable threads and apps the agent creates on its own.It is free for now, has no settled business model, and the company says it is in talks to raise more funding.
Circle and Proof demonstrated a flow in which an AI agent presents a verified credential over Proof's x401 identity protocol and pays in USDC over x402 from a Circle Agent Wallet.The pitch is to give merchants a way to verify who an agent acts for and get paid, rather than blocking agents outright.
OpenCode, an open source AI coding agent, said attackers broke into an older system and exposed API keys for 1.4% of users on its OpenCode Console platform, with some developers reporting hundreds or thousands of dollars in unauthorized charges.Anthropic and Google blocked the keys and OpenAI is issuing refunds, while the incident fueled debate over safety guardrails that stop defenders from analyzing malicious code.
MiniMax released M3.1 Flash on Monday, saying the model cuts agent costs by 70 percent while halving per token prices.Independent testing shows a smaller reduction in cost per completed task, because the real bill depends on how many tokens and loops an agent needs to finish.
MyDrivers reports that DeepBlue Technology has launched DFAgent, an agent operating system for film and TV production, and completed a signing for a film project.The company is positioning the product as China's first large model agent for the theatrical feature segment, though key details such as the partner and contract terms remain undisclosed.
BYD has launched an AI agent product called Didixia, framing the release as the company entering the AI agent era.The public information covers only the launch itself, with product form, capability boundaries and target vehicles still undisclosed.
Beyond Now has released Wave AI Framework, a product aimed at getting multiple AI agents to collaborate across the revenue lifecycle of communications service providers.StreetInsider reports the launch, which pushes agentic automation from standalone tools into the back office systems that determine how telecom revenue is booked.
A 36Kr report says Anthropic has all but stopped hiring junior staff, with the share of junior roles falling 33% in 18 months.The same report says work hours handled by agents now run six times those of employees.
Yelp is embedding a set of new AI features into its app, including one that phones a restaurant on a user's behalf to ask about current wait times.The company is also adding personalized recommendations that remember a diner's preferences, moving AI from answering questions to performing concrete tasks inside the app.
DataGroomr introduced an AI agent that lets Salesforce teams manage data quality through natural language, according to a PR Newswire release.The announcement did not disclose pricing, availability or the models behind the agent, leaving its practical scope unclear.
Tencent has released Hy Memory, an open source model it says leads the open source field on benchmarks it defined itself, with Claude Opus 5 as the comparison point.Because the results are vendor reported, the claimed margin still depends on third party reproduction and real developer adoption.
DeepSeek 新模型登顶开源模型榜首；国内 55% 算力用于视频和短剧？豆包发言人辟谣；微信发起公众号小说改编微短剧激励｜极客早知道 极客公园.DeepSeek 新模型登顶开源模型榜首；国内 55% 算力用于视频和短剧？豆包发言人辟谣；微信发起公众号小说改编微短剧激励｜极客早知道 极客公园
Memory Tensor released and open sourced MemT 1.0 on October 20, a memory native large model that moves memory out of external retrieval and into the model's own architecture, learning when to read, write and update it.Its technical report says MemT 1.
DeepSeek has published DeepInfer on GitHub, the high performance inference framework it uses to serve its models at scale, with the release focused on offline inference.The code covers optimized kernels, parallelism strategies, deterministic sampling and structured outputs, giving outside teams their first direct look at how DeepSeek actually runs its models.
Hugging Face shipped AnyLanguageModel 1.0 alongside 1.0 releases of swift huggingface and swift xet.Watch real task performance, API pricing, context capability, and migration cost.
Tolan, an AI companion app for managing daily life and emotions, raised $30 million from Union Square Ventures, Khosla Ventures and Homebrew, doubling its total funding.The startup says it has more than 250,000 active users, 70 million conversations in the past year, and annual recurring revenue in the double digit millions.
Tenstorrent has raised $1.5 billion at a $6.2 billion valuation, more than doubling the $2.Watch compute supply, deployment cost, inference throughput, and model service pricing.
German Chancellor Friedrich Merz and NVIDIA CEO Jensen Huang discussed Germany's industrial AI roadmap in Berlin, the same day Berlin signed a joint declaration of intent with NVIDIA on physical AI.Huang also announced planned investments in four German startups and a Siemens partnership on autonomous factories.
Microsoft introduced Guardian Agents and other security features designed to monitor the AI agents enterprises already use, framing them as potential insider threats.Its security chief said frontier models cannot be trusted regardless of origin, arguing that oversight must sit outside the labs that supply the models.
Cognizant is putting 18,000 staff behind Google's Gemini as part of a push into AI agents, according to The Next Web.The report does not detail specific workflows or efficiency targets, but the scale shows how IT services firms have become a distribution channel for frontier models.
An AI agent deployed in Centalion's trade finance amendment process has cut handling time by 80%, according to PYMNTS.The result points to agents moving from pilots into document heavy back office work, where throughput rather than text generation is the metric that matters.
【今日主题前瞻】大厂密集发布Agent产品，智能体赛道竞争白热化 市场测评 市场 中金在线.【今日主题前瞻】大厂密集发布Agent产品，智能体赛道竞争白热化 市场测评 市场 中金在线 Watch planning quality, tool use, permission control, and result verification.
OpenAI is speeding up a new funding round targeting about $30 billion, aiming to have the money committed before Anthropic's initial public offering, Reuters reported, citing people familiar with the matter.Sources described the contest as a two horse race, while the report did not set out the investor list or the valuation terms of the round.
The Barcelona Supercomputing Center's AI Factory Accelerator Programme is now fully operational, offering startups sovereign compute, technical support and training.NVIDIA said some BSC AI Factory services became available through its Inception portal on October 20, with a MareNostrum 5 extension set to widen access across Europe.
Otter.ai announced on Tuesday that it has raised $320 million from General Catalyst's Customer Value Fund, a vehicle that earns returns from marketing results instead of diluting equity in a conventional round.The company says the money will go to enterprise growth, go to market, international expansion, and product innovation.
Employee Benefit News reports on an AI agent built for employee benefits brokers, designed to absorb administrative renewal work so advisers can spend more time with clients.The article frames the problem bluntly: employers are leaving their brokers, and paperwork heavy renewals are a big part of why.
Altrio has launched an AI agent that scans thousands of listings to find the ones that fit a customer's investment criteria.The product targets the first pass screening stage of investment research, but its market coverage, data sources and accuracy have not been disclosed.
Eka Robotics raised a $150 million Series A at a $1.45 billion post money valuation, led by a16z and F Prime.The MIT spinout pairs a force sensitive gripper with a vision force action model that it says picks about 1,500 objects an hour with better than 95% reliability.
The Department of Homeland Security ordered US civilian agencies to remove DeepSeek products and DeepSeek based third party services from their IT systems, and barred buying them.DHS cited data flows to China, possible malware and intelligence risks, extending a wider Trump administration crackdown on Chinese technology.
Embedded AI evaluators won’t be an easy solution, Hugging Face’s policy chief says.AI labs have staked their reputations on model benchmarking — the supposedly neutral and accurate system of evaluations that are designed to measure whether one AI product is better than another.However, the methods behind these evaluations are complicated an.
New York Mayor Zohran Mamdani told a Manhattan audience on Monday night that AI's spread is not inevitable and that residents should demand accountability from AI companies.Speaking at the People's AI Assembly with actor Ilana Glazer, he criticized federal inaction on AI regulation and pointed to his own city level school ban and AI disclosure rules for housing listings.
Anthropic is pitching lawmakers a bill that would let the government block an AI model only after it causes more than $1 billion in damage or 50 deaths, according to a report by Pasquale Pillitteri.The proposal replaces capability based oversight with an extremely high harm threshold, turning the frontier AI debate into an argument over two concrete numbers.
Startup Fortune reported that Tencent is leasing 100,000 AI chips from Oracle as a way around US restrictions on chip sales to China.The arrangement replaces direct hardware purchases with rented compute, though chip model, deployment location, and contract terms are not addressed.
AWS released TorchNeuron on Monday, a tool that lets PyTorch applications run on its Trainium chip after a one line configuration change instead of a code rewrite.Amazon is also pushing changes upstream to make PyTorch less tied to Nvidia hardware, a direct challenge to CUDA's long standing software moat.
Anthropic and AI infrastructure provider Nscale are reportedly close to announcing another $45 billion compute deal, roughly a month after signing one of the same size.Bloomberg's sources caution that talks are ongoing and the terms could still change.
Dataconomy reported on October 20 that DeepSeek is working with Huawei on AI chip design, shaping future silicon around its own models so training depends less on Nvidia.The account rests on unnamed sources and has not been publicly confirmed by either company.
Samsung is expanding its sprawling semiconductor complex in Pyeongtaek, South Korea, to meet memory demand from generative AI.Three fabs already run around the clock, a fourth is coming online, and construction of a fifth has restarted after the 2024 downturn, as Samsung tries to close the gap with SK Hynix in high bandwidth memory.
DeepSeek and MiniMax each released new Flash series models within 72 hours of one another, according to a 36Kr report that frames the pair as challengers to flagship tier generative AI.The compressed rollout window puts fresh attention on whether lighter product lines can now close the gap with top end models.
Claude新模型曝光！能跳出思维定势，修Bug治标又治本.OpenAI只掏出一个6.1 Terra… Watch real task performance, API pricing, context capability, and migration cost.
OpenAI CEO Sam Altman presented employees with three scenarios for keeping advanced AI under human control, according to a Chosun Ilbo report citing The Information.The options range from an IAEA style global regulator to a CERN like joint lab — and a last resort plan to let the first superintelligence dominate, which Altman himself called 'terrible.'
Meshy has launched Meshy Edit, a tool built on a new in house model that lets users lasso part of a 3D model and describe the change in one sentence.It returns a new model in about a minute while keeping everything outside the selection untouched, turning 3D iteration into targeted edits rather than full regeneration.
Filing images for Saidou's first AI branded car have been published, revealing a near production exterior and confirming CATL as the battery supplier.Pricing so far is only an outside estimate built from listed components, which puts the car above 200,000 yuan.
Dataconomy reported on October 20 that OpenAI's GPT 6.1 Sol scores 97.9% on the company's COBOL migration evaluation, up from 86.