Hy3, a language model from Tencent, processed 7.7 trillion tokens on the OpenRouter platform in less than three weeks. This previously unknown LLM dominated the popularity rankings, leaving the competition far behind. Max Woolf described this phenomenon on his blog as one of the strangest moments in the history of AI model rankings.
TL;DR: Tencent’s Hy3 achieved 7.7 trillion tokens on OpenRouter in under three weeks, dominating the rankings with a large margin over Claude, Gemini, and DeepSeek. The model costs just $0.14 per million input tokens. The data comes from Max Woolf’s analysis on minimaxir.com.
Why does Hy3 LLM dominate the OpenRouter rankings?
Hy3 generated 7.7 trillion tokens (7.7T) on the OpenRouter platform in a period of less than three weeks. This result was described by Max Woolf in his analysis on minimaxir.com. The model appeared in the rankings almost out of nowhere and immediately took the lead, leaving solutions from Anthropic, Google, and DeepSeek far behind.
Tencent, the Chinese technology giant, stands behind this model. Hy3 had not been widely discussed in the AI community before. Meanwhile, its usage on OpenRouter grew at a rate that surprised analysts. Woolf points out that this model does not just lead - it does so with a margin described as “large.”
Hy3 costs approximately $0.14 per million input tokens. This aggressive pricing drives adoption. The OpenRouter platform aggregates traffic from multiple providers, so a model with such a low price naturally attracts the attention of developers looking for cheap options.
The dominance of a single model on an aggregation platform is a signal that the LLM market is still very dynamic. The question is: how long can Tencent maintain this position?
Where did the mysterious model from Tencent come from?
Tencent, the creator of Hy3, is one of the largest technology conglomerates in China. The company has years of experience in developing language models, but Hy3 appeared on OpenRouter without a typical marketing campaign. The lack of prior buzz makes its sudden success even more noticeable.
The Spanish-language service El Ecosistema Startup confirms that Hy3 processed 7.7 trillion tokens in under three weeks. This corroborates data from Woolf’s analysis. The model debuted on the platform and almost immediately began generating enormous traffic.
The lack of detailed technical documentation adds to Hy3’s air of mystery. Tencent has not published an extensive research paper or architecture details. Developers use the model based on empirical results rather than formal benchmarks.
Low price is a key piece of this puzzle. It attracts developers who need large text processing volumes at minimal cost.
How does Hy3 compare to Claude, Gemini, and DeepSeek?
Comparing Hy3 with models from Anthropic (Claude), Google (Gemini), or DeepSeek requires looking at pricing metrics. Hy3 costs a fraction of what the competition charges. For example, Google’s Flash models recently became three times more expensive, as Stephan Miller notes. This makes a cheap alternative even more attractive.
Here is a breakdown of factors contributing to Hy3’s popularity:
- Price: $0.14 per million input tokens, the cheapest on OpenRouter
- Volume: 7.7T tokens in less than three weeks
- Origin: Tencent, one of the largest Chinese technology conglomerates
- Availability: exclusively via API on OpenRouter, no free web interface
- Dominance: leading the rankings with a “large margin” according to Max Woolf
- No marketing: the model appeared without a prior promotional campaign
- Mystery: no official technical documentation
- Speed: immediate adoption without prior publicity
The table below shows the pricing context on OpenRouter:
| Model | Price per 1M input tokens | Note |
|---|---|---|
| Hy3 (Tencent) | ~$0.14 | Cheapest on the platform |
| Google Flash | Increased price | Became 3x more expensive |
| Claude (Anthropic) | Significantly higher | Premium pricing |
| DeepSeek | Low, but higher than Hy3 | Popular alternative |
The most important thing is that Hy3 does not compete on quality - it competes on price. On a platform like OpenRouter, where many developers seek cost optimization, the cheapest model naturally attracts the most traffic.
What does Hy3’s success mean for the AI model market?
Hy3’s success on OpenRouter shows that the language model market is price-sensitive. This is a signal for the entire industry: there is huge demand for cheap AI infrastructure.
Stephan Miller writes plainly in his analysis: the cheapest model on the internet wins. This phenomenon has implications for companies like Anthropic or Google that maintain higher prices. If Tencent can offer similar utility for $0.14, price pressure across the entire market will increase.
Similarly to other mysterious low-cost models, Hy3 gains traction without typical announcements. This is another example of this pattern in the LLM market.
For developers, it is also a reminder that training your own LLM from scratch is an option, but using ready-made, cheap APIs can be more cost-effective. The market offers an increasing number of choices, and price is becoming the decisive factor for many production applications.
Platforms like OpenRouter serve as a visible test for model popularity. Rankings on this platform reflect the real choices of thousands of developers, not just marketing declarations from AI companies.
Is Hy3 the beginning of a price war among LLM providers?
At the same time, Google raised the prices of Flash models threefold, which Stephan Miller describes on his blog. These two events suggest a divergence in the LLM market.
On one hand, Tencent is aggressively lowering prices. Meanwhile, American providers are raising their rates. This contrast shows that the market is still searching for its price equilibrium. The open OpenRouter platform means developers can immediately respond to cost changes.
Moreover, cheap models do not need to be the smartest. They just need to be good enough for specific tasks such as text classification or data extraction. Hy3 likely hits exactly this niche. The low price compensates for any potential quality shortcomings.
The LLM market is entering a phase where operational cost is becoming more important than marginal improvements in response quality.
Why doesn’t the lack of Hy3 documentation deter developers?
The model processed 7.7 trillion tokens in under three weeks. This is proof that the lack of paperwork does not block adoption.
Developers test models empirically. They send queries, check results, measure costs. If a model works acceptably and costs a fraction of the competition’s price, the lack of documentation becomes secondary. The OpenRouter platform facilitates such testing by offering a unified API.
Furthermore, many production applications do not require the most intelligent model. Automation, content tagging, generating simple summaries - for these tasks, a model with basic competencies suffices. Hy3 most likely hits these needs.
This demonstrates the scale of interest that emerged without any marketing effort from Tencent.
What are the potential risks of using Hy3?
Every model without documentation and history carries operational risks. Tencent could change pricing, restrict access, or modify the model’s behavior at any time. The lack of clear terms of use makes it difficult to assess whether data sent to the API is secure.
Here are the main risk areas associated with using Hy3:
- Price stability: Tencent could raise the price from $0.14 without warning, following Google’s move
- Data privacy: no clear policy on whether queries are used for training
- Availability: the model could disappear from OpenRouter just as mysteriously as it appeared
- Consistency: without versioning guarantees, responses could change drastically from day to day
- Support: no official communication channels with the team responsible for the model
- Dependency: tying critical processes to a single provider
- Auditing: no way to verify how the model makes decisions
- Regulations: unclear compliance status with GDPR and other data protection regulations
Therefore, developers should treat Hy3 as a temporary solution requiring monitoring. It is worth comparing this with the approach described in the article about switching from ChatGPT to a local LLM, where independence from external APIs is a priority.
Nevertheless, the low price attracts projects with limited budgets. Risk becomes acceptable when operational costs drop by orders of magnitude. However, it is worth maintaining a contingency plan and not tying critical processes to a single, undocumented model.
Frequently Asked Questions
How much does Hy3 cost on OpenRouter?
Hy3 costs approximately $0.14 per million input tokens, making it the cheapest model on the OpenRouter platform. For comparison, Google’s Flash models recently became three times more expensive, as Stephan Miller notes on his blog.
Who created the Hy3 model?
Hy3 was created by Tencent, a Chinese technology conglomerate. The service El Ecosistema Startup confirms that the model processed 7.7 trillion tokens in less than three weeks on the OpenRouter platform.
Why is Hy3 so popular despite lacking documentation?
Low price drives adoption. Stephan Miller points out on his blog that the cheapest model on the internet wins, and Google raised Flash prices threefold. Developers choose Hy3 for tasks that do not require the highest quality.
Is Hy3 safe to use in production?
The lack of technical documentation and privacy policy poses a risk. Tencent has not published details about data processing, as Max Woolf confirms in his analysis on minimaxir.com. It is recommended to avoid sending sensitive data and to maintain a contingency plan.
Summary
Tencent’s Hy3 shows that the LLM market is entering a phase of price war. The model achieved 7.7 trillion tokens on OpenRouter in less than three weeks, dominating the rankings. The low price of $0.14 per million input tokens proved more important than the lack of documentation or publicity.
Hy3’s success suggests that for many production applications, cost is more important than marginal improvements in model intelligence. Google raised Flash prices threefold, which further pushes developers toward cheaper alternatives.
The lack of documentation and a clear privacy policy are risks to keep in mind. Hy3 is suitable for tasks that do not require high quality, but it should not be the only tool in critical processes.
If you are interested in building your own AI solutions, check out the article on how to train your own LLM from scratch. For those seeking independence from external APIs, it is worth reading about how to stop paying for ChatGPT and switch to a local LLM. Both paths provide greater control over costs and data privacy.