{"slug": "8-of-openrouters-top-10-most-used-ai-models-this-week-are-chinese-heres-why-that", "title": "8 of OpenRouter’s Top 10 Most-Used AI Models This Week Are Chinese — Here’s Why That Matters", "summary": "Chinese AI models now dominate OpenRouter's weekly usage rankings, with 8 of the top 10 models by token usage being Chinese, including Tencent's Hy4 Preview at #1 with 20.4 trillion tokens. These eight models account for roughly 78% of the token volume among the top 10, signaling a broader shift in developer adoption beyond just DeepSeek.", "body_md": "Something interesting is happening in the AI model market.\n\nLooking at this week's **OpenRouter model usage ranking**, 8 of the Top 10 models by token usage are Chinese models.\n\nYes — **8 out of 10.**\n\nHere is the current ranking:\n\n| Rank | Model | Weekly Tokens | \n|---|---|---|\n| 1 | Tencent: Hy4 Preview | **20.4T** | \n| 2 | OpenAI: GPT-5.6 Luna | **14.8T** | \n| 3 | DeepSeek: DeepSeek V4 Flash 0731 | **12.9T** | \n| 4 | Z.ai: GLM-5.3 Flash | **12.7T** | \n| 5 | DeepSeek: DeepSeek V4 Flash 0423 | **5.15T** | \n| 6 | Tencent: Hy3 | **4.01T** | \n| 7 | NVIDIA: Nemotron 3 Ultra (free) | **3.82T** | \n| 8 | Xiaomi: MiMo-V2.5 | **3.82T** | \n| 9 | Z.ai: GLM-5.3 | **3.48T** | \n| 10 | Z.ai: GLM-5.2 | **2.51T** | \n\nThe only two non-Chinese models in the Top 10 are:\n\nEverything else comes from:\n\nAnd there is another number that makes this even more interesting.\n\nIf we add up the weekly usage shown in the ranking, the Top 10 models account for approximately:\n\n```\n83.59 trillion tokens\n```\n\nThe eight Chinese models account for approximately:\n\n```\n64.97 trillion tokens\n```\n\nThat's roughly:\n\n**78% of all token usage represented by the Top 10.**\n\nSo Chinese models aren't just occupying many positions on the leaderboard.\n\nThey're also representing the majority of actual token volume within this Top 10.\n\nThat's a much more interesting signal.\n\nPerhaps the biggest surprise is the model at the top.\n\n**Tencent Hunyuan Hy4 Preview** currently sits at #1 with:\n\n**20.4 trillion tokens**\n\nThat's significantly ahead of OpenAI GPT-5.6 Luna at 14.8T.\n\nHy4 Preview is Tencent's latest large MoE model, with:\n\nWhat's interesting is that Hy4 isn't positioned as a simple chatbot model.\n\nIt's designed around a broader shift:\n\n```\nQuestion\n   ↓\nAnswer\n```\n\nis becoming:\n\n```\nGoal\n ↓\nPlan\n ↓\nUse tools\n ↓\nExecute\n ↓\nObserve\n ↓\nAdjust\n ↓\nContinue\n ↓\nComplete task\n```\n\nThat kind of workload can also consume a lot of tokens, which is worth remembering when interpreting usage rankings.\n\nDeepSeek occupies two positions in the Top 5:\n\n```\n12.9T tokens\n5.15T tokens\n```\n\nCombined, that's more than:\n\n**18 trillion tokens in one week**\n\nacross these two V4 Flash versions alone.\n\nDeepSeek has become one of the most recognizable Chinese AI brands globally, but this ranking shows something more important than awareness:\n\n**developers are actually using the models at scale.**\n\nAnd the Flash family illustrates an important trend in AI infrastructure.\n\nNot every workload needs the largest and most expensive frontier model.\n\nFor many production applications, developers care about a balance between:\n\n```\nIntelligence\n     ×\nSpeed\n     ×\nCost\n     ×\nReliability\n```\n\nA model that is slightly weaker on a benchmark but dramatically cheaper to run can be much more attractive at scale.\n\nZ.ai may actually have the most interesting representation in this ranking.\n\nThree GLM models appear in the Top 10:\n\nGLM-5.3 Flash alone processed:\n\n**12.7T tokens**\n\nThat puts it almost level with DeepSeek V4 Flash 0731.\n\nThe recent GLM direction is also interesting because the models are becoming increasingly focused on:\n\nGLM-5.3 Flash is particularly notable because it combines:\n\n```\n320B total parameters\n18B activated parameters\n```\n\nwith a highly efficient hybrid attention architecture.\n\nThis is another trend we're seeing across Chinese AI labs:\n\n**The race isn't only about making models bigger. It's also about making intelligence cheaper to run.**\n\nA year ago, when many international developers talked about Chinese AI, the conversation often started and ended with:\n\nDeepSeek.\n\nThat picture is changing quickly.\n\nLook at this Top 10 again:\n\n```\nTencent\nDeepSeek\nZ.ai\nXiaomi\n```\n\nFour different Chinese companies are represented.\n\nAnd other major Chinese AI ecosystems include:\n\nThis matters because we're not looking at one breakout model anymore.\n\nWe're looking at an increasingly broad **AI model ecosystem**.\n\nDifferent companies are competing on different dimensions:\n\n```\nCoding\nReasoning\nAgents\nMultimodal\nLong context\nLatency\nCost\nTool use\nProductivity\n```\n\nThat creates much more choice for developers.\n\nThere is an important caveat.\n\nThis is a **usage ranking**, not an intelligence leaderboard.\n\n20T tokens does not mean a model is objectively better than one processing 5T tokens.\n\nToken usage can be influenced by many factors:\n\nFor example, an agent that reads an entire repository and performs dozens of tool calls can consume far more tokens than a chatbot answering simple questions.\n\nA cheap model may also be used much more aggressively than an expensive frontier model.\n\nSo the conclusion shouldn't be:\n\n\"Chinese models are better because they use more tokens.\"\n\nThe more interesting conclusion is:\n\n**Chinese models have clearly moved from being alternative models to models developers are actively using at very large scale.**\n\nThat's an important distinction.\n\nOne thing Chinese AI providers have been especially aggressive about is API pricing.\n\nThis changes how developers think about model selection.\n\nSuppose one model is 5% better for your workload but costs 10× more.\n\nWhich one should you use?\n\nFor a low-volume application, perhaps the stronger model.\n\nFor billions of tokens of production traffic, the answer may be very different.\n\nThis is why I think the most useful model metric is increasingly not:\n\n```\nBenchmark score\n```\n\nor even:\n\n```\nPrice per million tokens\n```\n\nbut:\n\n**Cost per successfully completed task**\n\nA coding agent might use a more expensive model but finish in fewer attempts.\n\nA cheaper model might consume more tokens but still have a lower total cost.\n\nThe economics depend on the workload.\n\nAnother thing that stands out in this ranking is the popularity of **Flash-class models**.\n\nWe have:\n\nnear the very top.\n\nThat makes sense.\n\nProduction AI workloads often need:\n\n```\nGood enough intelligence\n        +\nLow latency\n        +\nLow cost\n        +\nHigh throughput\n```\n\nrather than maximum benchmark performance on every request.\n\nFor example, an application might use:\n\n```\nSimple tasks\n     ↓\nFlash model\n\nMedium tasks\n     ↓\nGeneral model\n\nDifficult tasks\n     ↓\nFrontier reasoning model\n```\n\nThat's likely to become a common architecture.\n\nThe question isn't necessarily:\n\nWhich model should my application use?\n\nIt may instead become:\n\n**Which model should handle this particular request?**\n\nThis ranking is also another reminder of how quickly the model market changes.\n\nToday:\n\n```\nHy4 Preview\nDeepSeek V4\nGLM-5.3\nGPT-5.6\nMiMo\n```\n\nare receiving huge amounts of traffic.\n\nNext month, the ranking may look completely different.\n\nThat's why I think tightly coupling an application to a single AI provider is becoming increasingly limiting.\n\nA more flexible architecture could look like:\n\n```\nApplication\n     ↓\nAI Model Layer\n     ↓\n ├── GPT\n ├── Claude\n ├── Gemini\n ├── DeepSeek\n ├── Qwen\n ├── GLM\n ├── Hunyuan\n ├── MiniMax\n └── Others\n```\n\nThen select models according to:\n\nThe more competitive the model market becomes, the more valuable this flexibility becomes.\n\nOne of the things we're trying to solve with **ApiHub** is making this growing model ecosystem easier for developers to access.\n\nSeveral of the Chinese models appearing in this ranking are already available through ApiHub, including models from:\n\nInstead of creating a completely separate integration every time you want to test another model, ApiHub provides a unified way to access and compare different models.\n\nWe support multiple integration styles:\n\nAnd we provide **free credits** so developers can test different models before deciding what works best for their application.\n\nThe goal isn't to tell developers:\n\n\"Model X is the best.\"\n\nIt's to make it easier to answer:\n\n**\"Which model is best for my workload?\"**\n\nby actually testing them.\n\nThe headline is surprising:\n\n**8 of OpenRouter's Top 10 most-used AI models this week are Chinese.**\n\nBut I think the bigger story is underneath it.\n\nChinese AI models are increasingly competing on:\n\nAnd developers are actually using them.\n\nAt the same time, OpenAI, Anthropic, Google, NVIDIA, and other labs continue pushing the frontier forward.\n\nThat's good for developers.\n\nMore competition means:\n\n```\nMore models\n   ↓\nMore choices\n   ↓\nLower prices\n   ↓\nFaster iteration\n   ↓\nBetter AI applications\n```\n\nWe're moving away from a world where choosing an AI model meant choosing between only two or three companies.\n\nThe model layer is becoming a competitive marketplace.\n\nAnd that may ultimately matter much more than who happens to be #1 this week.\n\nWhat do you think?\n\n**Are Chinese AI models becoming part of your default model stack, or do you still mainly use GPT, Claude, and Gemini?**\n\nAnd if you've tried Hy4, DeepSeek, GLM, Qwen, or other Chinese models, which one surprised you the most?\n\n**Disclosure:** I'm building **ApiHub**, a unified AI API platform designed to make multiple AI models easier to access, test, compare, and integrate.", "url": "https://wpnews.pro/news/8-of-openrouters-top-10-most-used-ai-models-this-week-are-chinese-heres-why-that", "canonical_source": "https://dev.to/apihub/8-of-openrouters-top-10-most-used-ai-models-this-week-are-chinese-heres-why-that-matters-35ad", "published_at": "2026-09-09 08:05:56+00:00", "updated_at": "2026-09-09 08:30:57.066757+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-infrastructure"], "entities": ["OpenRouter", "Tencent", "DeepSeek", "Z.ai", "Xiaomi", "OpenAI", "NVIDIA", "GLM"], "alternates": {"html": "https://wpnews.pro/news/8-of-openrouters-top-10-most-used-ai-models-this-week-are-chinese-heres-why-that", "markdown": "https://wpnews.pro/news/8-of-openrouters-top-10-most-used-ai-models-this-week-are-chinese-heres-why-that.md", "text": "https://wpnews.pro/news/8-of-openrouters-top-10-most-used-ai-models-this-week-are-chinese-heres-why-that.txt", "jsonld": "https://wpnews.pro/news/8-of-openrouters-top-10-most-used-ai-models-this-week-are-chinese-heres-why-that.jsonld"}}