Wednesday, 28 October 2026
Google DeepMind's Genie 4 Makes Playable Worlds, Now With Sound
Google DeepMind’s Genie 4 Makes Playable Worlds, Now With Sound
Google DeepMind released Genie 4 on Tuesday, its newest “world model”: you describe a place, and it generates a scene you can walk through in real time with the keyboard, changing it as you go by typing — rain, a moved camera, an extra person. The new part is sound. DeepMind calls it “our first real-time world model to generate audio natively,” meaning ambient noise, music and characters’ dialogue are made together with the picture, not added afterwards. It also claims better physics, collisions and aerodynamics, and characters that react more naturally. It is built on Gemini Omni, and for now it is a research preview for trusted testers, coming “soon” to Google AI Ultra subscribers through Project Genie and to early partners on the Gemini API. DeepMind lists its own weak spots: objects still drift and change over longer runs, text in the scene is not reliably legible, and long-term consistency is unsolved. A world you can walk into, which forgets where it put the furniture.
Anthropic Offers Claude to Every Level of US Government at 40% Off
Anthropic widened its deal with the US General Services Administration on Tuesday: a 40% discount on Claude Enterprise, Claude for Government and the Claude API, open not only to federal agencies but to state, local, tribal and territorial governments, with no minimum spend — agencies pay for what they use. Anthropic estimates the discount will save the government $1 billion; it does not say for how long the offer runs. The announcement carries usage figures too: about 500 US government bodies use Claude, and weekly Claude for Government users are up roughly eightfold since February. One example is a Utah pilot where 34 state developers got Claude Code, and 71% said they saved 20 or more hours in four weeks. All of those numbers are Anthropic’s. Selling to a government at 40% off is still selling to a government — the kind of customer that rarely leaves once the paperwork is done.
Source: Anthropic, October 27, 2026
Anthropic Reports Early Drug Work With Six Partners — All of It Before Human Trials
Anthropic published a progress report on Tuesday from six drug-industry partners using Claude, plus its own program. The figures are the companies’ own and every one is early. At Generate Biomedicines, Claude proposed a part for a CAR T-cell therapy about 70% shorter than the standard one; in the lab, immune cells carrying it made roughly ten times as many receptors. Divergence Labs had Claude produce about 35,000 antibody designs in a month for food allergies; three of them beat the standard medicine at one lab task. BigHat tested about 1,010 Claude-guided variants in ten weeks, and the best bound more than ten times tighter. At Enveda, Claude went through 1,061 pages of manufacturing documents, caught all 11 problems human reviewers had found plus seven they had missed, and cut review time from over 25 hours to under three. Bristol Myers Squibb and The Column Group gave no numbers. Anthropic’s own muscular-dystrophy work, from its acquisition of Mithril Medicine, has promising results in mice, and could reach human trials “as soon as late 2028.” The report says it plainly: “All of this research is still at an early stage.” The paperwork is the part that already works.
Source: Anthropic, October 27, 2026