The Arena

Unisound's agent business clears 5B yuan in H1, token revenue jumps 500%

Editorial

Editorial

— 4 min read
Share
Unisound's agent business clears 5B yuan in H1, token revenue jumps 500%

Two Chinese AI stories landed today that point in opposite directions on what "working" looks like at scale: a Hong Kong-listed industry-agent company posted real growth on a Palantir-style model, and one of the country's largest healthcare AI products inked a deal to plug authoritative Western medical journals into its doctor-facing app.

Hong Kong's "AGI first stock" delivered a clean H1: revenue of 5.62 billion yuan, up 38.7% year over year, with the agent business alone accounting for roughly 85% of the total. Unisound (云知声) reported on Thursday that its enterprise intelligent services line — built from agent applications, an agent platform, and integrated solutions — hit 4.78 billion yuan in the first six months of 2026, up 35.7%. The company is the first Chinese AI firm to be marketed as an "AGI stock" in Hong Kong, and the H1 numbers are the first hard test of whether that framing maps to actual business. By the most important metric — whether the model is making money, not just spending it — the answer so far is yes. Gross margin rose to 33.1% as gross profit jumped 42% (faster than revenue), and the net loss rate narrowed by more than 31 percentage points, all while R&D spend climbed 69% to 284 million yuan. Re-purchase revenue already accounts for more than 60% of the total, the strongest signal in the report that the customers are staying.

The model underneath the numbers is the U2 family, a sparse mixture-of-experts stack with 260 billion total parameters activating roughly 10 billion per inference, which Unisound has wrapped in a medical-grade U2-Med layer plus voice and document multimodal variants. The new piece of the report is the token business — a small line in absolute terms (about 30 million yuan in H1), but with a Q2 sequential jump north of 500% and a gross margin above 60%, driven by dollar-denominated revenue and what the company calls a "high-value scenarios" strategy rather than competing on Token price. Cash on hand ballooned 302.8% to roughly 1 billion yuan, and the order backlog crossed 1.5 billion yuan, mostly from medical institutions — 470+ served in total, more than 80% of them tier-3 hospitals. The reason this story matters beyond Unisound is the structural shape: revenue and gross profit rising while R&D and backlog rise in tandem, and the loss narrowing, is the rare combination Chinese AI investors are looking for. The first Chinese labs to claim repeatable agent revenue are starting to look like software businesses with attached models, not model businesses hoping to add software later. We covered MiniMax's revenue nearly quadrupling as its model lags last week, and the H1 reports this season are quietly rewriting what the bar for "working" looks like in Chinese AI.

Ant Health plugs Wolters Kluwer's 28 top medical journals into its doctor app

Ant Group's healthcare arm, Ant Health, has signed a strategic deal with Wolters Kluwer to bring 28 of the medical publisher's flagship journals — including the Journal of Clinical Oncology and Annals of Surgery — into the Ant Afu doctor-facing app as the first phase of a deeper clinical-AI partnership. The journals cover 17 second-level medical disciplines and update in lockstep with the global editions, so doctors in China get evidence parity with their US and European peers on day one rather than waiting for translated or abridged reprints. The integration is the most concrete signal yet that China's consumer-facing medical AI is moving from "patient self-triage" into the physician's actual workflow — Afu's doctor version launched in August with an "AI medical assistant" and an "AI avatar" tool, and it already sits on top of 60 million evidence-based articles, medical guidelines, and 140,000 drug monographs.

The deal goes wider than journal access. Ant Health says clinician feedback from inside the app will flow back to Wolters Kluwer's UpToDate engine to refine its AI capabilities, with separate model tuning for tier-3 specialists, general practitioners, and primary-care doctors. The framing from Christian Cella, Wolters Kluwer's vice president for international medical markets, is the most honest version of why a Western evidence authority is interested in a Chinese AI partner: China's patient volume per doctor is so high that physician time is the binding constraint, and patients using AI for self-diagnosis paradoxically add to primary-care workload. Afu's "patient-side + doctor-side" loop, the argument goes, lets AI absorb the low-value triage work and lets humans keep the high-value clinical time — the same labor-economics pitch that drove the original wave of US clinical AI deals, just a decade and a regulatory framework behind. We covered earlier this month how China's industry ministry is laying out its five-year AI application playbook, and medical-AI deployment is one of the few areas where Beijing, US publishers, and Chinese tech are pulling in the same direction.

If the first wave of Chinese AI labs reporting earnings is finally showing the Palantir-shaped revenue curve, are the US open-weight labs next? Tell us in the comments.

Read more

AI 101 — How do AI video generators work?

AI 101 — How do AI video generators work?

An AI video generator is a model that starts with a grid of random noise and repeatedly refines it into a moving picture that matches your prompt — the same trick image generators use, stretched across time. Nothing is filmed, and nothing is stitched together from clips the model has seen. The pixels in the finished video were invented, frame by frame, all at once. Why it matters right now Kling 4.0, Kuaishou's flagship video model, went fully live on October 15 with up to 30 seconds of vid

Houmo moves compute inside memory — and A20 enters production

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

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

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