Tencent’s hy4 AI soars, fuels china’s model surge

The gist

Tencent’s Hy4 and ultra-budget Hy4 Flash are shaking up China’s AI scene, slashing enterprise costs and catapulting Chinese models to the top of OpenRouter’s charts almost overnight.

What to know

  • Tencent launched Hy4 and low-cost Hy4 Flash in early November 2026, with Hy4 preview touting 770B parameters and a record 1M-token context window.
  • Hy4 Flash undercut rivals by 95%, offering multimodal AI at just $0.07 per million tokens for common office tasks like research and slide decks.
  • Within days of launch, Hy4 soared to No. 2 on OpenRouter, while DeepSeek’s Flash models grabbed 35% of all platform usage, signaling a major Chinese AI surge.

Inside Tencent’s Launch Window

Hy4’s debut was telegraphed by a trail of pre-release signals and immediate investor reactions, revealing Tencent’s strategic orchestration behind the AI rollout.

The launch window is visible first in Tencent’s own pre-release signaling: TechNode reported on October 29 that the “upcoming Hy4 and Hy4 flash models will support visual understanding,” establishing that both variants were imminent ahead of the November 4–17 period. That matters because the August 20 strategy discussion cannot carry this chapter’s claim: it contains no mention of “Hy4” or “Hy4 Flash,” no dates inside the November 4–17 window, and instead discusses “Honeywan 3” and “Honeyuan 4,” with management saying “Honeyuan 3 is currently succeeding on cost performance.” Likewise, Tencent’s later Q3 2026 earnings call on November 12 falls within the November 4–17, 2026 window but is retrospective and does not itself announce the launches, so it cannot directly support the chapter’s claim about those specific releases driving adoption and investor response.

By early November, the launch had become concrete and market-visible: South China Morning Post reported on November 4 that Tencent “Launches Hy4 AI,” while TradingView said Tencent “released Hy4” and noted the stock gave back some of the prior day’s 5.93% gain after the Hy4 launch, evidence of immediate investor response. The enterprise angle also surfaced in the November 1 ad-hoc-news.de headline tying the unveiling to WorkBuddy “serving 50% of China’s Top 500 Enterprises,” whereas the August transcript offers no Hy4-linked adoption proof and only says analysts were “sweating over these strategic priorities.”

Sources

Enterprise AI at Unmatched Scale

Hy4 Flash’s architecture and pricing slashed costs for businesses, enabling seamless integration into office workflows and fueling a surge in enterprise AI adoption.

Tencent had already laid the integration groundwork before Flash arrived: TencentHunyuan released “Hy4-preview with 770B total / 49B active and 1M context,” explicitly calling it “open source frontier,” while kimmonismus highlighted Tencent’s claim that “Hy4 can coordinate multiple Codex sessions in parallel for research workflows.” That mattered because the model was not just large but deployable: vllm_project described a serving stack with “256 routed experts + 1 shared,” only “21/78 layers” computing sparse indices while others reused them, plus an “embedded 10B MTP layer with draft depth 3,” all pointing to enterprise-ready throughput.

Flash turned those capabilities into a scale product by collapsing cost and targeting routine enterprise usage: TechNode’s headline described Hy4 Flash as a “Multimodal AI Model” that “Slashes Costs 95% to $0.07 / M Tokens,” explicitly tying its positioning to cost reduction, and Pasquale Pillitteri described it as a 1M-context model built for “everyday office work” such as research, documents, and slide decks. Because Tencent paired multimodal and agent-style inheritance with a low-cost, long-context deployment profile, enterprises could plug it into existing workflows faster, raising usage intensity and token volume growth while strengthening Tencent’s competitive position in China’s AI market.

Sources

Chinese Models Dominate Platforms

A tidal wave of domestic demand pushed Hy4 and DeepSeek to the top of OpenRouter, cementing China’s rapid ascent as a global AI powerhouse.

The mid-November OpenRouter rankings did not emerge in a vacuum; they reflected a Chinese AI market already expanding at extraordinary scale. Official Chinese statistics showed daily AI token consumption rising from approximately 100 billion at the beginning of 2024 to 100 trillion by the end of 2025, then exceeding 140 trillion by March 2026, while commercial demand was visible in August 2026 when Alibaba said its AI Cloud and Compute Services business generated about $7.1 billion in quarterly revenue; earlier, Hugging Face estimated Chinese-developed models accounted for roughly 41% of model downloads over the preceding year.

By the week of November 9, that demand was clearly translating into platform traffic concentration around Chinese providers. OfficeChai reported that “Tencent’s new Hy4 has leapt straight to second place,” with OpenRouter listing the model as released on November 4, meaning it reached the No. 2 slot within days, while DeepSeek remained the anchor of that shift: “Add its two Flash models together and DeepSeek accounts for 70T tokens, or 35% of everything on the platform,” making Hy4’s rapid ascent further evidence that domestic models were taking commanding share.

Sources

Get the stories behind the trends

Deep-dive reporting and the weekly brief, in your inbox.