OpenRouter 排行榜顯示 token‑maxxing 已停滯
Token-maxxing is dead, OpenRouter's token usage plateaus
OpenRouter 最新排行榜顯示 token‑maxxing 趨勢已停滯，DeepSeek V4.1 Flash 為主流模型，Google 與 OpenAI 佔高使用量。
LLM Rankings | OpenRouter
[](https://openrouter.ai/workspaces/default)
Alpha
Search⌘K
ModelsBenchmarksChatChatRankingsAppsOriPricingDocs
Sign Up
Sign Up
Top ModelsLeaderboardTop models by taskCost per sessionMarket ShareOpen / ClosedBenchmarksFastest modelsLanguagesProgrammingDomainContext LengthTool CallsImagesTop Apps
Top Models
AI Model Rankings
Live LLM rankings by real-world usage. Models are ranked by tokens processed through the OpenRouter API by millions of developers.View all text models
Usage data through Oct 18, 2026
Top Models
Weekly usage of models across OpenRouter
Model rankings by tokens processed
| Rank | Model | Author | tokens processed | Change |
| --- | --- | --- | --- | --- |
| 1 | DeepSeek V4.1 Flash | deepseek | 32.8T tokens | -20% |
| 2 | GLM 5.3 Flash | z-ai | 12.2T tokens | 0% |
| 3 | Step 5 Preview | stepfun | 11.5T tokens | +30% |
| 4 | MiMo-V2.6-Flash | xiaomi | 9.94T tokens | -13% |
| 5 | Hy4 preview | tencent | 7.55T tokens | -9% |
| 6 | GPT-6 Luna | openai | 6.21T tokens | -12% |
| 7 | Claude Opus 5.5 | anthropic | 6.18T tokens | +23% |
| 8 | GLM 5.3 | z-ai | 5.19T tokens | +37% |
| 9 | Nemotron 3 Ultra (free) | nvidia | 4.97T tokens | -28% |
| 10 | DeepSeek V4 Flash 0731 | deepseek | 4.7T tokens | -4% |
1.
bydeepseek
32.8T tokens
20%
2.
byz-ai
12.2T tokens
0%
3.
bystepfun
11.5T tokens
30%
4.
byxiaomi
9.94T tokens
13%
5.
bytencent
7.55T tokens
9%
6.
byopenai
6.21T tokens
12%
7.
6.18T tokens
23%
8.
byz-ai
5.19T tokens
37%
9.
bynvidia
4.97T tokens
28%
10.
bydeepseek
4.7T tokens
4%
Show more
Top models today, as text
Text summary of the Today tab of the leaderboard above. Models are ranked by tokens processed on OpenRouter over the most recent complete day, with the change measured against the day before it.
Models ranked by tokens processed on OpenRouter over the most recent complete day
| Rank | Model | Author | Tokens | Change |
| --- | --- | --- | --- | --- |
| 1 | DeepSeek V4.1 Flash 1017 | deepseek | 3.04T | +345% |
| 2 | DeepSeek V4.1 Flash | deepseek | 2.62T | -36% |
| 3 | GLM 5.3 Flash | z-ai | 1.43T | -7% |
| 4 | MiMo-V2.6-Flash | xiaomi | 1.04T | -10% |
| 5 | Hy4 preview | tencent | 846B | -7% |
| 6 | Claude Opus 5.5 | anthropic | 822B | +3% |
| 7 | Nemotron 3 Ultra (free) | nvidia | 790B | >999% |
| 8 | GPT-6 Luna | openai | 644B | -2% |
| 9 | DeepSeek V4 Flash 0731 | deepseek | 608B | +4% |
| 10 | GLM 5.3 | z-ai | 525B | -17% |
Top models this month, as text
Text summary of the This Month tab of the leaderboard above. Models are ranked by tokens processed on OpenRouter over the trailing thirty days, with the change measured against the thirty days before them.
Models ranked by tokens processed on OpenRouter over the trailing thirty days
| Rank | Model | Author | Tokens | Change |
| --- | --- | --- | --- | --- |
| 1 | DeepSeek V4.1 Flash | deepseek | 124T | +582% |
| 2 | Space Bunny Alpha | stealth | 56.7T | new |
| 3 | GLM 5.3 Flash | z-ai | 54.8T | +30% |
| 4 | MiMo-V2.6-Flash | xiaomi | 37.8T | new |
| 5 | Hy4 preview | tencent | 35.3T | -22% |
| 6 | DeepSeek V4 Flash 0731 | deepseek | 25.2T | -49% |
| 7 | Nemotron 3 Ultra (free) | nvidia | 24.7T | +34% |
| 8 | GPT-6 Luna | openai | 23.2T | new |
| 9 | Step 5 Preview | stepfun | 20.3T | new |
| 10 | GPT-5.6 Luna | openai | 17.9T | -63% |
New and trending models this week, as text
Text summary of the New & Trending tab of the leaderboard above. Models are ranked by the percentage change in tokens processed over the trailing seven days against the seven days before, among models with at least one million tokens in the current week. A model with no prior week is marked new.
Models with the largest week-over-week change in tokens processed on OpenRouter
| Rank | Model | Author | Tokens | Change |
| --- | --- | --- | --- | --- |
| 1 | DeepSeek V4.1 Flash 1017 | deepseek | 3.72T | new |
| 2 | Nemotron 3.5 Super VL (free) | nvidia | 805B | new |
| 3 | Gemini 4 Argon | google | 767B | new |
| 4 | Jev 1.14 | typesafe | 596B | new |
| 5 | Muse Wonder | meta | 233B | new |
| 6 | Classify-1 | scaledown | 7.81B | >999% |
| 7 | GPT-5.5 (batch) | openai | 24.8M | >999% |
| 8 | Claude Sonnet 5 (batch) | anthropic | 7.91M | >999% |
| 9 | Gemini 2.5 Pro (batch) | google | 82.4M | >999% |
| 10 | GPT-5.6 Sol Pro (batch) | openai | 14.5M | >999% |
Top models by task
Each task’s leading models, ranked by share of spend on OpenRouter
Share of spend
Cost per session
What one coding-agent session typically costs (paid usage), by session length
Market Share
Absolute Percentage
Compare text request share by model author on OpenRouter.
Market share by model author, as text
Text summary of the Market Share chart above. Share of text requests made on OpenRouter in the week beginning Oct 12, 2026, with the change in each author's requests against the week before it. DeepSeek leads at 24.4% of the week's requests. Authors the chart groups as others are listed last, so the rows match the series the chart plots.
Model authors ranked by their share of text requests on OpenRouter in the most recent complete week
| Rank | Author | Share of requests | Change in requests |
| --- | --- | --- | --- |
| 1 | DeepSeek | 24.4% | +3% |
| 2 | Google | 19.7% | +14% |
| 3 | OpenAI | 18.5% | +3% |
| 4 | Z.ai | 12.2% | -6% |
| 5 | Qwen | 5.3% | +1% |
| 6 | Anthropic | 4.7% | +25% |
| 7 | Xiaomi | 2.4% | -14% |
| 8 | Meta Llama | 2.1% | |
| 9 | Mistral AI | 2.0% | +5% |
| | All other authors | 8.6% | -9% |
Open / Closed
Absolute Percentage
Share of tokens processed by open-weight vs. closed-weight models on OpenRouter, by week. Stealth models are excluded.
Benchmarks
Top models on OpenRouter by Artificial Analysis Intelligence Index
Weighted Avg Input Price Intelligence Index Score
- [x] Show Pareto
TOP 20 PLOTTED SEARCH TO PIN UP TO 5 Find a model…
1.
Claude Opus 5.5 (Max, Default Fallback)
57.6
2.
Claude Sonnet 5.5 (Max, Default Fallback)
56.0
3.
Claude Fable 5.1 (Max, Default Fallback)
53.4
4.
byqwen
53.4
5.
byopenai
52.7
6.
bygoogle
52.6
7.
byopenai
51.8
8.
50.8
9.
Claude Fable 5 (Max, Opus 4.8 Fallback)
49.6
10.
bymeta
48.1
Show more
Domain
Compare models by the subject area of requests on OpenRouter
Top models by context length
Summary of the context-length chart above for the 10K context window in the week beginning Oct 12, 2026. Others leads with 2b requests.
Models ranked by requests in the most recent complete week
| Rank | Model | Requests |
| --- | --- | --- |
| 1 | Others | 2b |
| 2 | typesafe/jev-1.13-20260917 | 610m |
| 3 | z-ai/glm-5.3-flash-20260826 | 380m |
| 4 | deepseek/deepseek-v4.1-flash-20260910 | 291m |
| 5 | openai/gpt-6-luna-20260922 | 223m |
| 6 | deepseek/deepseek-v4-flash-20260423 | 219m |
| 7 | deepseek/deepseek-v4-flash-20260731 | 192m |
| 8 | google/gemini-2.5-flash-lite | 150m |
| 9 | openai/gpt-oss-120b | 104m |
| 10 | typesafe/jev-1.14-20261014 | 89m |
Tool Calls
Absolute Percentage
Tool usage across models on OpenRouter
Images
Absolute Percentage
Total images processed on OpenRouter
Top Apps
Today
Explore apps and agentsopting into usage trackingby category
Browse the new Apps & Agents directory to explore by category, discover trending tools, and see detailed usage stats.
1.
Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessions, and builds reusable skills from experience. It comes with 40+ built-in tools, including web search, browser automation, and vision, plus scheduled automations and subagents.
1.39T tokens
2.
Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, runs tests, and iterates on failures, all from natural language prompts.
1.38T tokens
3.
Kilo Code is an open-source AI coding agent that works across VS Code, JetBrains, and CLI to help developers ship code faster with agentic workflows.
559B tokens
4.
Cline is an open-source AI coding agent that lives inside your IDE, autonomously exploring your codebase, editing files, running terminal commands, and using browser automation.
366B tokens
5.
There are many coding agents, but this one is yours.
355B tokens
6.
new
329B tokens
7.
Powerful coding models, funded by ads
322B tokens
8.
new
247B tokens
9.
new
223B tokens
10.
OpenClaw is an open-source AI agent that connects to your messaging apps and takes real actions on your behalf, from running commands and browsing the web to managing files and sending emails.
179B tokens
How these rankings are measured
What is measured
Each model is ranked by the number of tokens it processed through the OpenRouter API, counting both prompt and completion tokens. Usage is aggregated into daily buckets in UTC and totalled per model variant, so variants of the same model, such as a free variant, are ranked separately. Requests that a user or app keeps private are excluded before aggregation.
Selection period
Today, This Week, and This Month each cover a trailing window of one day, seven days, and thirty days, ending with the most recent complete daily bucket. Trending compares the trailing seven days with the seven days before it and ranks models by the percentage change in tokens, including only models with at least one million tokens in the current window so that a small base cannot produce a large percentage. New models with no prior week are listed first, up to five of them.
Freshness
The date above the rankings is the newest usage bucket the leaderboard is computed from, not the time the page was rendered or deployed, so it advances only when new usage data is aggregated.
How to read these rankings
These rankings show traffic routed through OpenRouter. The Top Models chart and the LLM Leaderboard rank models by tokens processed, while the Market Share chart counts requests. None of them measures users or spend. Models differ in verbosity and tokenization, so a higher token total shows how much a model is used, not which model is best for a task. They do not rank models by accuracy, reasoning ability, or benchmark performance. The Benchmarks section ranks those using scores from Artificial Analysis and Design Arena, plus benchmarks such as GPQA Diamond that OpenRouter runs itself and reports as median accuracy across providers. For coding, compare the Artificial Analysis Coding Index there with the Programming section, which ranks models by tokens on a sample of requests in a chosen language, such as Python. The Top Models chart names the nine most-used models in each period and groups the rest as Others. The Market Share chart shows each model author's share of text requests on OpenRouter, with the remaining authors grouped as Others.
Rankings data by OpenRouter is licensed underCC BY 4.0. Reuse and republish it with attribution. TheData API docscover the citation format and JSON access to these datasets.
Footer
© 2026 OpenRouter, Inc
Product
- Chat
- Rankings
- Benchmarks
- Apps
- Discover
- Models
- Ori
- Collections
- Providers
- Tools
- Pricing
- Business
- EU routing
- Enterprise
- Labs