Hy3 preview from Tencent has surged to the top of OpenRouter's model rankings, outperforming even DeepSeek V4 Flash in usage despite lower benchmark scores. OpenRouter data shows Hy3 receives over 98% of input tokens in API calls, suggesting cost-driven adoption. Its stated price of $0.066/1M tokens input is 34% cheaper than DeepSeek V4 Flash's $0.10/1M rate, though effective pricing varies by provider due to cache efficiencies. Tencent's model, while not surpassing Claude or GPT in quality, benefits from a free-to-paid transition in May that built organic traction.

The mystery deepens as Hy3's popularity isn't tied to major apps or agentic workflows. OpenRouter logs reveal less than 1% of activity comes from top applications, yet usage remains steady since its paid launch. This contrasts with past ranking spikes caused by free offerings like Grok Code Fast 1. The model's single provider—Singapore-based SiliconFlow—has seen minimal prior interest, making Hy3's rise appear anomalous. OpenRouter's effective pricing table shows DeepSeek V4 Flash can drop as low as $0.018/1M tokens via direct caching, but Hy3's $0.066 rate still undercuts most providers. This discrepancy hints at either aggressive pricing strategy or user preference for Tencent's ecosystem.

The key takeaway isn't Hy3's technical superiority but its cost advantage in a market where LLM expenses are skyrocketing. OpenRouter's data confirms 98% of API costs stem from input tokens, which are aggressively cached—reducing real-world expenses. Hy3's affordability likely attracts price-sensitive users, even if its performance lags. However, this model challenges the assumption that lower costs equate to inferior quality. Whether Hy3's success signals a shift toward value-driven LLMs or a temporary anomaly remains unclear. OpenRouter's transparent ranking system provides rare insight into this dynamic, but the long-term implications for LLM economics are yet to unfold.