Honor ships the first phone with a system-level agent harness

Two stories from the Chinese AI beat this morning, one about where the agent runs and one about what it can simulate.

Honor unveiled MagicOS 11 today with what it calls the industry's first commercially deployed system-level Agent Harness, and the numbers behind it are bigger than the label. Its YOYO assistant now strings together more than 100 steps in a single instruction, with roughly 700 system tools it can call — nearly double the previous set — and more than 500 third-party Skills already published. Honor's own benchmark framing is that it closes about 9 of 10 complex multi-app tasks end to end, from booking a flight to scheduling a reminder two days before departure. The harness matters more than the assistant: perception, planning and execution live in the OS rather than inside an app wrapper, so the agent can act across apps the way a user does, and Honor plans an AgenticOS preview on the Magic9 series in October. That is the same bet this site has been tracking all day — the model is moving onto the hardware, both in silicon (MediaTek's 2nm chip puts a 30B MoE model on your phone) and in policy (China's new five-year plan puts the AI model on the device). Honor is first to ship it as a shipping product rather than a demo.

MoleculeMind's QuantaMind force field simulated a 17,792-atom enzyme doing its job for 6 nanoseconds, and the paper landed in Science Advances. Most AI molecular models learn from stable-state snapshots, so they fall apart the moment a bond breaks; QuantaMind was trained on 5,286 reaction transition points and kept its total-energy drift small enough for continuous runs measured in hundreds of nanoseconds. For the PETase test — the plastic-degrading enzyme, plus substrate and nearly 5,000 water molecules — it reproduced proton transfer, bond breaking and the full catalytic cycle without a pre-scripted reaction path, and random configurations spot-checked at quantum-mechanical level agreed above 0.99. The company says the same setup now reaches 100,000 atoms on a single GPU, which is a claim about cost, not just capability: reactive simulation has been supercomputer work, and dropping it to one card puts it inside normal R&D budgets. The company's founder Jinbo Xu built RaptorX-Contact, the first working AI structure-prediction method and an acknowledged ancestor of AlphaFold, so this team has shifted paradigms once already.

What to watch: whether Honor opens the harness to non-Chinese app ecosystems, since an agent is only as useful as the services it can reach.

If your phone could run 100 steps on your behalf, which task would you hand over first — travel, money, or email? Tell us in the comments.

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Read more

Houmo moves compute inside memory — and A20 enters production

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Brain organoids classified Braille at 83% — in a trio, not alone

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute