On the night of April 11, Beijing’s E-Town district hosted a full dress rehearsal for what is now becoming an annual spectacle: a humanoid robot half marathon. More than 70 teams showed up for the overnight test run, navigating the complete 21-kilometre course through urban streets and ecological park terrain, with engineers and support staff following on electric carts and motorcycles just to keep pace.

The official race takes place on April 19, running alongside a human half marathon on the same route. Over 100 teams are registered this year, nearly five times the number that competed in last year’s inaugural event. Four international teams are among the participants. The robots run in two categories: autonomous navigation, where the machine finds its own way around the course, and remote-controlled, where humans guide them.

Autonomous teams now make up about 40% of the field, a significant jump that organizers describe as one of the defining challenges of this year’s race.
Robots taller than 75 centimetres are required to complete the full distance in a single continuous effort, and each robot will wear a BeiDou Navigation Satellite System-powered shoulder badge for centimetre-level positioning and real-time location tracking throughout the race.

The speed numbers are getting genuinely impressive. Unitree Robotics announced a new world record in humanoid robot sprint speed during the lead-up to the event, with its H1 robot reaching around 10 metres per second, approaching the peak human sprint speed achieved by Usain Bolt during his 100-metre world record. Some teams are now predicting their robots could finish the half marathon at times close to elite human runners.

The event is being watched as more than just a competition. Organizers describe it as a major step forward for the robotics industry, accelerating the transition of humanoid robots from laboratory prototypes to real-world applications. China’s humanoid robotics sector has been moving fast, from coordinated group performances at the Spring Festival Gala to new commercial deployments, and this race is shaping up to be one of the clearest demonstrations yet of just how far the technology has come.

Anthropic is currently in a legal battle with the Trump administration over the Pentagon’s decision to label the company a supply chain risk. At the same time, the company has been briefing that same government on Mythos, its most powerful and unreleased AI model. Those two things are both true, and Anthropic co-founder Jack Clark confirmed it publicly this week.

Speaking at the Semafor World Economy summit, Clark explained the company’s position clearly. Anthropic’s view is that the US government needs to know about powerful AI models regardless of what is happening in the courts, because the national security implications are too significant to let a legal dispute get in the way. Clark described the Pentagon dispute as a narrow contracting disagreement and said it should not overshadow the fact that Anthropic genuinely cares about national security. He said the company briefed the Trump administration on Mythos and will continue to do the same with future models.

The context around Mythos makes that decision feel significant. The model, previewed last week through a cybersecurity initiative involving over 40 major tech and security companies, has been described internally as Anthropic’s most powerful yet, capable of finding thousands of previously unknown software vulnerabilities in a matter of weeks. It is not being made publicly available precisely because of how capable it is. Reports have also suggested that Trump officials are encouraging major US banks including JPMorgan, Goldman Sachs, and Citigroup to test the model.

Clark also touched on broader concerns about AI and employment during the interview. While Anthropic CEO Dario Amodei has warned publicly about potential unemployment reaching Depression-era levels, Clark offered a slightly more measured read, saying the company is currently only seeing weakness in early graduate employment across select industries. He said Anthropic is monitoring the situation and prepared to respond if larger employment change materialise.

Google has added a new capability to Gemini’s Personal Intelligence feature that lets the AI generate images using personal context it has already gathered from your Google account, without you having to spell everything out in the prompt.

The way it works is that Gemini draws on information connected through your Gmail and Google Photos to understand your preferences, interests, and personal life. Instead of typing a detailed prompt like “generate an image of my dream home, I love tennis and music,” you can simply say “design my dream home” and Gemini fills in the rest from what it already knows about you. It can also pull from labels and names in your Google Photos, so prompts like “generate an image of my family doing our favourite activity” will actually produce something relevant rather than a generic output.

The feature includes a sources button that shows you exactly what context Gemini used to create the image, which is a useful transparency feature given how much personal data is feeding into the output. If Gemini gets something wrong, users can provide feedback or add reference photos manually by tapping the plus icon.

This sits within Google’s broader Personal Intelligence feature, which launched earlier this year and has been rolling out steadily. It became available to all US users in March and this week expanded to users in India and other countries. The image generation addition is powered by Nano Banana 2, Google’s advanced image model, and will roll out to Plus, Pro, and Ultra subscribers in the US in the coming days, with Gemini in Chrome desktop and broader availability to follow.

Anthropic has launched Claude Design, a new experimental product that lets users create visual outputs like prototypes, slide decks, one-pagers, and other visual assets directly through Claude. The product is specifically aimed at people who need to get from an idea to something visual quickly but do not have a design background, think founders, product managers, and anyone who has ever sent a rough sketch to a designer with the words “something like this.”

The workflow is straightforward. You describe what you want and Claude creates an initial version. From there you can refine it through direct edits or by giving instructions, adjusting colours, typography, layout, or asking for things like a dark mode toggle. The aim is to skip the blank page problem entirely and get to something usable fast.

One of the more useful enterprise features is that Claude Design can read a company’s codebase and design files to learn its design system, meaning every output it creates automatically stays consistent with the company’s existing visual style. Teams can maintain more than one design system and apply whichever is relevant to a given project.

Anthropic was careful to frame Claude Design as a complement to tools like Canva rather than a replacement. Once you have created something in Claude Design, you can export it as a PDF, a URL, a PPTX file, or send it directly to Canva where it becomes fully editable and collaborative.

The product is powered by Claude Opus 4.7 and is currently available as a research preview for Pro, Max, Team, and Enterprise subscribers. The launch sits in the context of a busy stretch for Anthropic, which has been rapidly expanding its workplace product lineup alongside recent news that investors have been offering preemptive funding at a valuation of $800 billion or more, an offer the company has reportedly been holding off on.

Building AGI with our mission Intelligence with Everyone. Global leader in multi-modal models and AI-native products with over 150 million users.