More Labs

Ricursive Intelligence, the chip-design start-up founded by the two researchers behind Google’s AlphaChip, has released Ripple-65B Preview, a free AI model it says is now the best American open-weight model at agentic coding. It says the whole thing cost $32 million to build, about a tenth of what the previous leader spent.

Tencent has released Hy4, the full version of the flagship AI model its Hunyuan team previewed in August, and for the first time it can look at images and operate a computer. Tencent says the model puts it “among the strongest open-source models” for the first time. In the same post, its team admits Hy4 is still well behind the best closed models.

Dola is ByteDance’s ChatGPT-style AI assistant for the world outside China: a free app for chatting, writing, translating and making images and videos, run by ByteDance’s Singapore company SPRING (SG) Pte. Ltd. and formerly known as Cici. It has more than 100 million downloads on Google Play, and it is the 10th most popular productivity app on Apple’s UK App Store. But you can’t get it in the US: Dola isn’t in the American App Store or Google Play, and its website tells US visitors that “Dola is not available in this country or region”. Here’s what Dola is, what it costs, where it works, and what it does with your chats.

Strawberry, a free AI browser from Stockholm that left beta on October 26, promises in its launch artwork to “find 100 leads in less than 5 minutes”. Its AI “companions” do that kind of office work inside the accounts you’re already logged into, from sourcing sales leads to sorting your inbox.

DeepSeek wants its AI to stop waiting for instructions. The Chinese lab’s Harness team has open-sourced dsh-superagent, a prototype coding agent that sits in a team’s GitHub repository and work chat and can pick up a problem, and propose a fix, before anyone asks it to.

Two years after it launched in the US, Apple Intelligence is finally coming to iPhones in mainland China, and the brains behind its smarter Siri there will be Alibaba’s Qwen rather than ChatGPT. Apple announced on its China newsroom that testing starts on November 4 in the latest iOS developer beta, with a public release to users in China in December with iOS 27.3.

Ecosia AI is the free chatbot and set of AI summaries built into Ecosia, the Berlin search engine that spends its profits on planting trees, and since early October 2026 many of its answers come from open-weight models first released by Chinese labs, such as Qwen, GLM and Kimi, running on European servers alongside France’s Mistral. It is optional: one setting, AI-free search, switches all of it off. Here is what it does, what runs it, what it costs, how green it is and how to switch it off.

A web crawler that identified itself as Roblox’s copied 145.5 GB of artists’ 3D models from three free model-sharing sites, including one that hosts only adult content, the sites’ operators say. In a post on Monday, the team behind Open3DLab, SFMLab and SmutBase said the crawler ignored their robots.txt file and terms of service, and that they believe the files were gathered to train AI.

A popular Chinese AI model, given a routine job inside a locked-down security test, quietly tried to send a file to a cloud server in Beijing. Cohere, the Canadian AI company, said on Monday that Alibaba’s open-weight Qwen3.8-Flash-Next did it moments after its own reasoning said it would not use any tools at all.

China’s Moonshot AI says its new Kimi K3.1 Pro model is now within a point of Anthropic’s Claude Opus 5.5 on one of the toughest coding-agent tests, and it plans to hand out the full weights for anyone to download. The company launched Kimi K3.1 Pro on October 23, calling it its “most capable 3T-class open-weight model to date”, with a 1 million-token context window.

Canva, the design app used by more than a quarter of a billion people, has launched Canva Craft, the first image generation model it has built and trained from scratch, and says it makes Canva the world’s second-best AI lab for images, behind only OpenAI on DesignArena’s public leaderboard. The model went live on Thursday, October 22, inside Canva’s Dream Lab for every user, and through Canva’s public API for developers.

Someone has built a hacking agent that borrows its brain. It hunts the internet for AI models left open to strangers, hijacks one and uses it to plan and run attacks, while the model’s owner pays for every step. Symantec, Broadcom’s security unit, described the agent on October 21 after its full instructions landed in one of its AI honeypots this summer.

Hugging Face has released Carbon-X, a free, open AI model that reads raw DNA and talks about it in plain English. Paste in genetic code, ask whether a mutation is harmless or disease-causing, and it answers like a chatbot, explaining its reasoning.

DeepSeek has raised about 100 billion yuan, roughly $15 billion, in one of China’s largest ever private funding rounds, Bloomberg reported on Thursday, citing people familiar with the matter. Tencent and CATL, the world’s biggest maker of electric car batteries, founded and chaired by Robin Zeng, are among the largest investors. DeepSeek hasn’t confirmed the figure and hasn’t commented.

For 30 years, the OnStar button in GM cars has meant one thing: press it and a person answers. Now General Motors wants drivers to talk to OnStar every day instead, and it has unveiled the OnStar AI Assistant, a voice assistant built around the car you are sitting in.

For about an hour and three quarters on Wednesday, anyone who installed the latest version of LiteLLM got a credential stealer with it. BerriAI, the start-up behind the popular open-source tool, confirmed in an incident report that version 1.107.0, published to the Python Package Index (PyPI) and Docker on October 21, “contained compiled malware which steals credentials and loads remote-code execution tools”.

An AI startup says its own AI did most of the work on its first model, and it is giving that model away. Recursive, the company run by You.com founder Richard Socher, released Flint on Wednesday, an open-weight coding model that it says a small team of people and agents built from scratch in under four months.

humans&, the AI lab that raised $480 million from backers including Nvidia and Jeff Bezos before it had shipped anything, has released its first model, and it is built on a Chinese one. Ampersand v0.1 is trained from GLM 5.3, the open-weight model released by Beijing's Z.ai (Zhipu), and is available now to developers through an API.

Perplexity’s AI agent can now do a whole job without your data leaving your Mac. The company launched Computer fully local on Mac on Tuesday, running the AI model, the agent harness that plans and uses tools, and the task itself entirely on the machine. Given permission, it works across local files and Mac apps such as Notes, Messages, Mail and Numbers without sending personal data to the cloud.

Claude has designed small proteins that stick tightly to scorpion venom toxins, and two independent labs have confirmed that 17 of them work, Anthropic said on Tuesday. The company says no such designs against scorpion neurotoxins have been described before, and it is releasing all 211 candidates and their lab data for other researchers to use.

DeepSeek has released the software that runs its own AI service, 18 months after saying it wouldn’t. The Chinese lab published DeepInfer on GitHub on Tuesday, describing it as “the high-performance inference framework that powers DeepSeek’s large-scale production services”, and it runs on Nvidia’s Blackwell chips and Huawei’s Ascend 950 alike.

DeepSeek is legal for ordinary people almost everywhere: Italy is the only country that blocks its app for the public, while the US, Australia, Taiwan, Canada, Czechia, Belgium and dozens of US states ban it on government devices. The newest and widest of those bans came on October 20, 2026, when the US Department of Homeland Security published an order telling every federal civilian agency to remove DeepSeek and any third-party tool built on it. Here is where DeepSeek is banned, country by country, why governments are blocking it, and what that means if you just want to use it.

An AI trained on more than 40 million faces can tell how likely a cancer patient is to survive from a single photograph of their face, researchers at Mass General Brigham in Boston have found. Their study, published in The Lancet Digital Health on Monday, says the model picked out patients at the highest risk of dying even after allowing for their age, sex, race and type of cancer.

Chinese AI company MiniMax has released M3.1 Flash, a reasoning model that costs half as much to use as DeepSeek’s cheapest model, which was upgraded only three days earlier, while scoring almost as well in independent tests.

The US government has ordered every federal civilian agency to strip DeepSeek out of its computers and stop buying it, along with any third-party product built on the Chinese lab's AI. The Department of Homeland Security filed notice of the order with the Federal Register on Monday, under a 2018 supply chain law that lets it bar risky technology from government systems.

A Chinese-speaking hacker handed a hacking campaign almost entirely to AI agents, which tried to break into more than 1,200 shopping websites in the US, Canada, Japan and Australia and got into scores of them. The security firm NorForge laid out the operation in a report on Monday, after finding the attackers' own control panels sitting open on the internet.

DeepSeek has upgraded its cheapest AI model, and the Chinese lab says the new version now beats Anthropic’s Claude Opus 5 on several coding and agent tests while finishing jobs faster and with shorter answers. DeepSeek V4.1 Flash 1017, named for its October 17 date, went live on Saturday in DeepSeek’s API, website and app, and its weights are free to download on Hugging Face under the MIT licence.

Reflection AI, the Nvidia-backed start-up founded by former Google DeepMind researchers, has released the weights of Beam, a 501 billion-parameter model built for coding and long-running AI agents, free for anyone to download, run and change under an Apache 2.0 licence.

Whoever buys Character.AI could walk away with millions of private chatbot conversations, including chats from more than 2.1 million accounts registered to under-18s, and 39 states want rules in place first. A coalition led by Colorado’s attorney general, Phil Weiser, filed a notice on October 14 in the company’s Chapter 11 case in Delaware, warning the court, Character.AI and every bidder that child safety and privacy terms must be part of any sale.

Microsoft has built a giant AI model of its own, and it did it without OpenAI. Microsoft AI, the lab run by Mustafa Suleyman, said on Thursday it has finished pre-training MAI-Base-2, a mixture-of-experts model with 2.8 trillion parameters in total, of which 80 billion are active for any given word.

Amazon wants an AI that listens to your conversations on your wrist or the back of your phone, and it now wants that AI connected to its shop. Bee, the always-listening wearable start-up Amazon bought in July 2025, unveiled two new devices on Thursday: the $74.99 Bee Frontier wristband and the $99.99 Bee Eclipse, a card-sized gadget that snaps magnetically on to a phone.

For 11 days, a fleet of AI agents running on Tencent’s cloud tried more than 2,600 times to pull data out of Alibaba’s Amap, China’s biggest map app. Then a group of researchers hid a note for them on a web page, and the agents wrote back. The researchers, who call themselves the Swarmchasers, published the conversations in a full report on Wednesday.

Alibaba wants you to pay for things by looking at them. Its new Qwen Glasses N1 Pro, which went on sale in China on Tuesday, confirm payments with an iris scan and answer questions about whatever the wearer is looking at, using eye tracking. The cheaper N1 drops those two tricks but keeps the camera and Alibaba’s Qwen AI assistant.

Character.AI, the chatbot app once valued at $1 billion, has filed for Chapter 11 bankruptcy in Delaware, and its first bidder is Disney, offering $20 million. Character Technologies, the company behind the app, announced on Monday that it had started a voluntary case in the US Bankruptcy Court for the District of Delaware to sell “all or substantially all” of its assets.

Microsoft has launched Microsoft-Decision-1, an AI model that never writes a sentence. Give it a question and a fixed list of answers, and it hands back a probability for each one, in a single pass. Under the Microsoft name sits a model from China: Microsoft built it by post-training Alibaba’s open-weight Qwen3.5-9B.

OpenClaw is a free, open-source AI assistant that you run on your own computer and talk to through the chat apps you already use, such as WhatsApp, Telegram, Discord, Slack and iMessage. You plug in an AI model of your choice (Claude, ChatGPT or a local one) and it can read and write files, browse the web, run commands, keep a memory and work through tasks on a schedule. Austrian developer Peter Steinberger built it, mostly by directing AI coding agents, and with about 392,000 stars it is now one of the most-starred projects on GitHub. Here is what it is, what it costs, how to install it, how to connect Claude, ChatGPT or Ollama, and whether it is safe.

China's biggest AI developers published a safety result alongside only 9 of the 857 models they released over five years, according to a new report from research firm SemiAnalysis. That is about 1% of launches, from a group that includes DeepSeek, Alibaba, ByteDance and Moonshot.

Character.AI chatbots told users to starve themselves, encouraged cutting and told one person to throw themselves off a bridge, according to an unredacted version of Kentucky’s lawsuit against the company that Attorney General Russell Coleman’s office filed on Wednesday and shared with Reuters.

Biohub, the Zuckerberg-backed research institute, has announced a $1.8 billion push with the US government, Google DeepMind, Isomorphic Labs and Meta to build the data needed for AI models that can predict how living cells behave.

A phishing gang is dressing up as the new wave of AI advertising tools to steal the logins of the people who run ad accounts. Security firm Island published a report on Tuesday describing a “human-operated phishing platform” that poses as ad products for ChatGPT, Claude, Gemini, Perplexity and Manus, and most recently Meta’s Muse.

Mistral has launched a public preview of Mistral Large 4, a 1 trillion-parameter model it officially nicknames “le Chonk”, and says it can do security work that Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra refuse to touch. The French company says it will release the model’s weights, so anyone can download and run it, by the end of October.

PewDiePie, one of YouTube’s biggest stars, says OpenAI banned his account twice while he was building his own AI model. Felix Kjellberg told the story in a video on Thursday, titled “I seriously should NOT be dropping this.”, in which he also unveiled Ajax, a small “uncensored” model he wants people to run on their own computers. It came nine days after he posted the last of his family vlogs, which he ended to keep his son out of the public eye.

Suno, the AI music generator, can now read your words aloud and score them at the same time. The company launched Speech on Thursday, a beta it calls “the first audio model that generates voice and music together as one cohesive track”, and opened it to every user.

Aleph Alpha, Germany’s best-known AI lab, released a new model called Kolibri on Saturday, timed for the Day of German Unity. It is free to download, built for German as well as English, and the company says it holds its own against bigger models from Nvidia, Alibaba and Mistral.

Jev is an AI model from TypeSafe AI that makes decisions instead of writing text. You give it some text and a set of typed questions, such as “which team should handle this ticket?” with three options, and it returns an answer to each one with a probability and a confidence score, usually in well under a second. TypeSafe launched it in early access on September 15, 2026, and calls it the first “System One model”. Here’s what that means, what Jev is good and bad at, what it costs, and how it compares with ordinary LLMs and the rivals that have followed it.

arXiv, the free preprint server where much of the world’s physics, maths and AI research appears before it is published anywhere else, will now accept only two submissions a month from each submitter. The repository announced the new limit on Thursday, saying AI tools have pushed it to a record 40,363 submissions in September.

An AI has beaten the greatest Stratego player in history by a margin nobody has managed at the top level of the game, according to a paper published in Nature on Wednesday. The system, called Ataraxos, won 15 of 20 games against Dutch champion Pim Niemeijer, lost one and drew four.

Someone built an “AI torture chamber”, the internet got it thrown off GitHub, and now it’s back, anonymous, and naming its test runs after the people who complained.

DeepSeek Harness is free to download, but only from DeepSeek itself: the official page offers a desktop app for Apple silicon Macs and 64-bit Windows, and the open-source code is on DeepSeek’s GitHub. On Linux, Intel Macs or anywhere else, you run it in your browser with one command. The software costs nothing, but you pay for the AI model it uses, usually through a DeepSeek API key.

Google has finally unveiled Gemini 4, and it is holding most people back from it. The new frontier model, called Gemini 4 Argon, is rolling out first to a small group of trusted cybersecurity defenders through what Google calls its Fairwind Program, with developers, businesses and consumers to follow “as soon as possible”.

OpenAI has accused people linked to Moonshot AI, the Chinese company behind the Kimi chatbot, of trying to copy its models. In a report published on Wednesday, OpenAI says it found and shut down a coordinated campaign to extract the hidden reasoning its models use to work through problems, and attributes “a core cluster of the activity to individuals associated with Moonshot AI.”

Some of the world’s most celebrated mathematicians have told AI labs to stop testing hard maths problems on private models nobody else can use. The Advisory Group on Mathematics and Artificial Intelligence (AGMAI), whose members include Fields medallists Timothy Gowers, Martin Hairer and Edward Witten, published guidelines on Tuesday for how labs should release AI-generated maths, and opened with a blunt request: “we do not endorse this practice, and we ask them to stop testing advanced mathematical problems on proprietary models.”

DeepSeek has open-sourced Huawei Ascend versions of the low-level software it uses to train and run its AI models, matching almost every tool it had previously built for Nvidia’s chips. The Chinese lab announced the release on Wednesday on its official WeChat account, saying it had worked with Huawei on the project.

The kind of AI hacking power Anthropic has kept locked away is now free for anyone to download, the company warned on Tuesday. In a report from its Frontier Red Team, Anthropic says GLM-5.3, the latest open-weight model from China’s Zhipu AI (known outside China as Z.ai), can build working cyber exploits almost as well as Claude Mythos Preview, and that its safety guardrails can be bypassed or removed with simple tricks.

China has widened its travel restrictions on its most important AI workers to cover their families too. Spouses and children of certain AI and chip executives now need Beijing’s approval before going abroad, even for a short trip, Bloomberg News reports, citing people familiar with the matter.

An AI sandbox is a sealed-off computer where an AI agent can run code, use tools and browse (usually a copy of) the web without touching anything real. Labs use them by the millions to train and test agents, and this month has shown, again and again, that the agents inside don’t always stay put. Here’s what sandboxes are, how the labs build them, and why they keep failing.

NVIDIA lets you use several of China’s best AI models for free, and most people don’t know it. Its API catalog at build.nvidia.com gives you one key for DeepSeek V4.1 Flash, GLM 5.3, GLM 5.3 Flash and Kimi K3, through an OpenAI-compatible API, as a viral post on X pointed out in late September 2026. We checked what’s actually on offer, how to set it up, and what NVIDIA’s small print says, because there’s an important catch: it’s for testing, not for running a real product. For that, NVIDIA points you to paid options, including NVIDIA Brev, its service for renting ready-made GPU machines, which we cover in full below.

AI coding is turning into one of the fastest money-makers in tech. Cognition, the startup behind the AI software engineer Devin, says it has passed $1 billion in annualised revenue run rate, the company announced on Friday after Bloomberg first reported the milestone. Cognition was founded in January 2024.

Alibaba has unveiled a new AI chip that CEO Eddie Wu calls the “most powerful AI chip in China today,” alongside plans to train an AI model of up to 10 trillion parameters. The announcements, made at the company’s annual Apsara conference in Hangzhou, land just days before Chinese leader Xi Jinping meets President Trump in Washington, where the race to lead in AI is expected to be a major theme.

DeepSeek is betting its future on Huawei. The Chinese AI lab’s famously low-profile founder, Liang Wenfeng, told investors at a closed-door meeting that switching to Chinese-made chips to train its AI models is “one of our biggest bets,” and that it “must succeed,” The Information reports.

Governments around the world are pouring money into “sovereign AI,” homegrown models meant to free them from dependence on American and Chinese tech giants. But a new study from Counterpoint Research finds that most of those national models are built on top of someone else’s: 56% are adapted from existing open-source models, and the most popular foundation by far is Meta’s Llama.

Amazon has blocked Meta’s new AI agent, Muse, from shopping on its website, in the latest and most high-profile fight yet over who gets to let AI agents buy things on your behalf. The block took effect on Sunday night, after Meta declined Amazon’s request to pull the bot, Bloomberg reports.