China AI - SiliconNoon (Page 24)

The Arena

China's LLM call volume tops US for 15 straight weeks

The week's usage data landed, and the curve keeps pointing one way: Chinese models now move more traffic than American ones — by a wide margin. The catch is in what the numbers actually measure. Chinese AI models have now out-called their US rivals for 15 consecutive weeks, with the top four most-used models on the OpenRouter aggregation platform all coming from China. For the week of August 3–9, the platform recorded 69 trillion tokens of total usage, up 21.5% week over week.

The Arena

Honor's YOYO Claw agent now runs on Zhipu's GLM-5.3

China's device makers are plugging frontier models straight into their agents, and Honor just became the first big name to ship Zhipu's new flagship. Honor's cross-device AI agent YOYO Claw now runs on Zhipu's GLM-5.3, the model the lab released Thursday — making Honor the first major device maker to adopt the new open-weights model.

The Arena

Bank of China's first 'token loan' ties credit to AI usage

China's banks are starting to lend against something with no physical form: the tokens an AI company burns through. Bank of China has reportedly issued the country's first compute "token loan," and the product signals where AI financing is headed — past the data center and into the application layer. Bank of China has extended what's reported as China's first compute "token loan," sizing credit lines for AI companies by how many tokens their products actually consume.

The Arena

Star 50 index up 29% in 2026 as China's AI trade hits 150x earnings

China's AI boom has a second front, and it's in Shanghai's STAR Market. The Financial Times is flagging a benchmark that keeps outrunning its US counterpart — and now trades at a multiple that makes the Nasdaq 100 look cheap. Shanghai's tech-focused Star 50 index is up 29% in 2026 and trades at a price-to-earnings ratio above 150 — more than four times the Nasdaq 100's roughly 35 — the Financial Times reports.

The Guardrails

China poised to lift travel ban on Manus founders

Beijing is unwinding the last restraints of the AI industry's most consequential forced deal reversal, while fresh capital flows into the businesses that test models and wire the data centers running them. China plans to lift its travel ban on Manus' co-founders, and CEO Xiao Hong intends to return to Singapore as the company nears the end of its separation from Meta, the Financial Times reports.

The Frontier

PerceptionBench: no frontier model breaks 60% on basic vision

Frontier models can hold their own on math and code, but Moonshot AI's new benchmark suggests most of them still can't reliably see — and what gets blamed as a reasoning failure may actually start at the eyeball. Moonshot AI, the lab behind Kimi, released PerceptionBench, a benchmark that isolates visual perception from reasoning — and no frontier model clears 60 percent accuracy on it.

The Guardrails

US tells 35 partners: pick Pax Silica or China's AI bloc

Washington is drawing a hard line under the global AI cold war: partner with the US-led Pax Silica initiative, or with China's rival framework — not both. The US plans to tell 35 partner countries they will be excluded from the US-led Pax Silica initiative if they join China's rival AI framework, according to a Reuters exclusive based on a letter and a source familiar with the matter.

The Frontier

Qwen open-sources Qwen3.8-27B that runs on consumer GPUs

Alibaba's Qwen delivered the open-source model the community kept asking for, and Meta's most expensive hire walked out the door — a Friday that splits cleanly into two stories. Alibaba's Qwen open-sourced Qwen3.8-27B on Friday under the permissive Apache 2.0 license — a 27-billion-parameter dense multimodal model built to run on consumer GPUs, free for anyone to download, deploy, and sell.

Quick Hits

Quick Hits — August 14, 2026

A midday catch-up: Anthropic lets Claude Code maintain its own software, Qwen's compact 27B sibling is hours from release, and Tencent's Hunyuan brain trust starts moving toward WeChat. Anthropic now runs daily maintenance on its own software with Claude Code — and 46 percent of the AI's pull requests are getting merged after human review. Over a few weeks, the tool generated 388 pull requests covering everything from crash fuzzing to dead-code removal across the company's apps.

The Arena

SiEngine's Tiangong 100 car AI chip enters mass production

China's in-car AI race just got a new economics story: a 7-nanometer automotive AI chip priced in the thousand-yuan range is now shipping at volume. SiEngine, the Wuhan-based automotive chipmaker, announced on August 12 that its self-developed Tiangong 100 AI accelerator has entered full mass production, with batch shipments now going out to automakers and Tier-1 suppliers.

The Arena

Apple trains its own China AI model with Alibaba's support

Apple is reportedly training its own large language model specifically for the Chinese market — a departure from its plan to lean on third-party models — and Alibaba is helping it do it. Meanwhile, China's top foundries just posted the strongest quarters in their histories as AI demand spills into every corner of the chip supply chain. Reuters reports that Apple has trained a large language model specifically for the China market, developed in partnership with Alibaba, which supported the…

The Frontier

Zhipu's GLM-5.3 ships with 'emergent' cyber capabilities

China's open-weights race just produced its most awkward win: Zhipu's GLM-5.3 landed Thursday with frontier-level coding and a cyber capability its own makers admit outgrew their expectations — and Moonshot AI is racing a closing window toward a Hong Kong listing. Zhipu released GLM-5.3 on Thursday, an open-weights flagship whose every gain comes from post-training — including an "emergent" cyber capability the company says developed faster than it expected.

See all