OpenRouter is a platform that allows access to various AI models via a unified API, and it supports a wide range of AI models, including Anthropic's Claude and Google's Gemini. In the usage ranking published by OpenRouter, Tencent's AI model ‘Hy3 preview,’ a major Chinese IT company, maintains a high ranking of 1st or 2nd place.
LLM Rankings | OpenRouter
Furthermore, in the OpenRouter rankings, Hy3 preview held the top spot from the week of April 27th to the week of May 11th, 2026.
The official X account for the Hy series also reported that the Hy3 preview reached number one on OpenRouter's weekly rankings.
Two weeks after release, Hy3 preview is #1 on @OpenRouter 's weekly leaderboard with 3.66T tokens processed, up 298% week-over-week.
#1 in overall usage, tool calls, and coding. 15.4% market share across all providers.🏆
Top apps running Hy3 preview: Hermes Agent, Claude Code,… pic.twitter.com/LjCwtZ9zOm
The Hy3 preview is a model that Tencent released as an open model in April 2026.
Hy3 preview is a mixed expert (MoE) model with a total of 295 billion parameters and 12 billion active parameters, and a context window of up to 250,000 tokens. It also features three inference modes that prioritize either latency or depth.
Tencent Hyre
https://hy.tencent.com/research/hy3
tencent/Hy3-preview · Hugging Face
https://huggingface.co/tencent/Hy3-preview
The graph below shows benchmark results published on the Hy3 preview's Hugging Face page. It shows that while the Hy3 preview's performance has significantly improved compared to the previous generation Hy2 model in terms of software engineering performance (SWE), terminal environment task capabilities (Terminal), internet information retrieval capabilities (BrowseComp), and wide-area information gathering and organization capabilities (WideSearch), it is still somewhat inferior to models like Claude.
OpenRouter offered a limited-time free preview of Hy3 starting May 6, 2026, and it was during this free period that it achieved the number one ranking in the OpenRouter rankings. However, the free period has already ended, and the ranking at the time of writing this article is based on usage during the paid period.
The new Hy3-Preview model from
Looking at the fields in which Hy3 preview is used, it spans a wide range, including academia, finance, health, legal affairs, marketing, and programming.
According to tests conducted by BuzzFeed's Senior Data Scientist Max Wolf, the performance of Hy3 preview was on par with other Chinese-made models and far inferior to models like Claude Opus 4.7 and GPT 5.5. However, the price of Hy3 preview on OpenRouter is $0.066 (approximately 11 yen) per million tokens, which is cheaper than DeepSeek V4 Flash's $0.10 (approximately 16 yen) per million tokens, so the lower price may be driving the adoption of Hy3 preview.
Wolf pointed out that while the token price for Hy3 preview is indeed low, if you cache the same prompt to reduce the number of input tokens, DeepSeek V4 Flash will be cheaper if you choose the right provider. He said that if this price advantage becomes widely known, it wouldn't be surprising to see a surge in the use of DeepSeek V4 Flash in a few weeks.
This issue has also been discussed on Hacker News, with one user pointing out that OpenRouter's ranking is based solely on the total number of tokens, and that large users who use a massive amount of tokens daily may be using Hy3 preview .
The mysterious Hy3 LLM is topping OpenRouter Model Rankings by a large margin | Hacker News
https://news.ycombinator.com/item?id=48317294
Related Posts:
<< Next
It has been found that a decline in pollinating insects is causing a decrease in farmers' income and a deterioration in their nutritional status.
Prev >>
The former owner is furious after $200,000 worth of LEGOs he consigned were stolen from the store, and YouTubers have picked up on the story, leading to a review bombardment, but there are many suspicious points.
in AI, Posted by log1h_ik