Tencent Hy3 calls spike 68x in first week, topping OpenRouter

Tencent's Hunyuan Hy3 large model generated 68 times more API calls than its predecessor in its first week, claiming the top spot on OpenRouter's global leaderboard and showing strong enterprise demand for open-weight AI models.

"Hy3's adoption curve has been steeper than any model we've previously released, driven by developers who need a customizable base they can fine-tune without vendor lock-in," a Tencent spokesperson said.

The 295-billion-parameter Mixture-of-Experts model activates only 21 billion parameters per token and ships under Apache 2.0 with a 256,000-token context window. It scores 78.0 on SWE-Bench Verified, a benchmark that tests an AI's ability to resolve real GitHub issues. On WorkBuddy, a platform where enterprises select custom models for agentic workflows, 60% of users picking a custom model chose Hy3 over available alternatives.

The adoption surge puts pressure on competitors such as Alibaba's Qwen and Baidu's ERNIE to match Hy3's open-weight flexibility and pricing. Tencent shares rose 2.7% on the day, with short selling accounting for 15.7% of turnover, suggesting some traders remain skeptical of the AI monetization timeline despite the strong product signal.

The 68x call volume compares with Hy2, the previous-generation model, and extends momentum from Hy3's preview period. Tencent made Hy3 freely available on OpenRouter through July 21, a promotional strategy that likely accelerated trial adoption. The open-weight release under Apache 2.0 removes a key barrier for enterprise developers who need to inspect, modify, and self-host the weights.

Hy3's Competitive Positioning

At 21 billion active parameters per token, Hy3 sits in a cost-efficiency sweet spot: large enough to handle complex reasoning and agentic tasks, small enough to run on modest hardware. Its 256,000-token context window supports the long-document analysis and multi-turn agentic workflows that enterprise customers increasingly demand. The model's 78.0 on SWE-Bench Verified is closely watched by companies building AI coding assistants.

Tencent's approach contrasts with the closed-model strategy of OpenAI and Anthropic, whose most capable models require API access with per-token pricing. By releasing weights under Apache 2.0, Tencent allows enterprises to deploy Hy3 on their own infrastructure, avoiding API latency and data-privacy concerns. The trade-off: Tencent forgoes direct API revenue in exchange for ecosystem lock-in and cloud services upsell.

What This Means for Investors

Tencent's AI push has been a key narrative for the stock, which trades at roughly 20 times forward earnings. The Hy3 adoption data provides the first concrete evidence that Tencent's open-weight strategy is gaining traction against both Chinese rivals and global players. If the 60% WorkBuddy share translates into sustained enterprise contracts, Tencent's cloud and AI services revenue — currently a small fraction of its total — could become a material growth driver.

The risk: promotional pricing and free trials can inflate early adoption numbers. Whether Hy3 retains users after the OpenRouter promotion ends on July 21 will be the next test.

This article is for informational purposes only and does not constitute investment advice.