AI Model Usage Rankings — OpenRouter
The most-used AI models this week, ranked by total tokens processed through OpenRouter.
Snapshot 2026-10-15 · updated weekly
Usage rankings answer a different question than benchmarks: not which model scores best, but which models developers actually run. This leaderboard aggregates the weekly window of OpenRouter traffic — total prompt and completion tokens plus request counts — for the top 40 models, refreshed weekly. Flash-class models dominate raw token volume because they power high-volume pipelines, while frontier models lead on requests from coding agents. Models we track in depth link to their full benchmark page.
| Rank | Change | Model | Vendor | Weekly tokens | Requests |
|---|---|---|---|---|---|
| 1 | ▲4% | Deepseek V4.1 Flash | DeepSeek | 37.06T | 775,523,907 |
| 2 | NEW | Step 5 Preview | Stepfun | 19.48T | 100,416,271 |
| 3 | ▼39% | GLM 5.3 Flash | Zhipu AI | 12.95T | 674,052,171 |
| 4 | ▲7% | Mimo V2.6 Flash | Xiaomi | 11.22T | 139,455,207 |
| 5 | ▲7% | Hy4 Preview | Tencent | 7.88T | 71,748,799 |
| 6 | ▲18% | Nemotron 3 Ultra 550b A55b | Nvidia | 7.31T | 57,464,498 |
| 7 | ▲3% | GPT 6 Luna | OpenAI | 6.81T | 379,557,620 |
| 8 | ▼59% | Claude Opus 5.5 | Anthropic | 5.85T | 47,186,790 |
| 9 | ▲46% | GLM 5.3 | Zhipu AI | 4.83T | 62,904,005 |
| 10 | ▼10% | Deepseek V4 Flash | DeepSeek | 4.73T | 333,846,706 |
| 11 | ▼1% | Jev 1.13 | Typesafe | 3.28T | 1,062,956,044 |
| 12 | ▲79% | Hy3 | Tencent | 3.26T | 41,320,580 |
| 13 | ▼2% | Deepseek V4 Flash | DeepSeek | 3.11T | 348,623,986 |
| 14 | ▲94% | Muse Spark 1.3 Contributor | Meta | 3.09T | 40,132,462 |
| 15 | ▲39% | Claude Sonnet 5.5 | Anthropic | 2.78T | 39,118,829 |
| 16 | ▲37% | Kimi K3 | Moonshotai | 2.53T | 38,774,497 |
| 17 | ▲46% | GPT 6.1 Sol | OpenAI | 2.29T | 51,819,236 |
| 18 | ▲421% | Claude Haiku 5.5 | Anthropic | 2.01T | 106,764,588 |
| 19 | ▼7% | Gemini 3.8 Flash | 1.95T | 85,017,095 | |
| 20 | ▼8% | GPT 5.6 Luna | OpenAI | 1.75T | 147,381,945 |
| 21 | ▲36% | Mimo V2.6 Pro | Xiaomi | 1.64T | 24,094,659 |
| 22 | ▲17% | Minimax M3 | Minimax | 1.39T | 25,386,165 |
| 23 | ▲1% | Nemotron 3.5 Lightning | Nvidia | 1.31T | 23,219,684 |
| 24 | ▼35% | GLM 5.2 | Zhipu AI | 1.09T | 49,919,636 |
| 25 | ▼2% | Laguna S 2.1 | Poolside | 1.08T | 13,821,974 |
| 26 | ▲33% | Gemini 3 Flash Preview | 1.08T | 79,321,419 | |
| 27 | ▼5% | Deepseek V4 Pro | DeepSeek | 1.03T | 51,189,569 |
| 28 | ▲19% | Solar Mini4 | Upstage | 989B | 11,160,828 |
| 29 | ▲30% | Solar Pro4 | Upstage | 829B | 12,411,089 |
| 30 | ▼45% | Claude Opus 5 | Anthropic | 795B | 7,751,718 |
| 31 | ▲44% | Dots 3 Note Preview | Dots Studio | 784B | 15,585,355 |
| 32 | ▼27% | Nemotron 3 Super 120b A12b | Nvidia | 754B | 19,551,187 |
| 33 | ▼62% | GPT 5.6 Sol | OpenAI | 747B | 23,604,669 |
| 34 | ▼8% | Deepseek V4 Pro | DeepSeek | 695B | 27,211,400 |
| 35 | ▼18% | Claude Sonnet 5 | Anthropic | 669B | 16,685,429 |
| 36 | ▼16% | Gemini 2.5 Flash Lite | 623B | 292,253,570 | |
| 37 | ▼71% | GPT 6 Sol | OpenAI | 590B | 15,575,209 |
| 38 | ▲23% | Qwen3.8 Flash | Qwen | 556B | 23,549,898 |
| 39 | ▲15% | Ling 3.1 Flash | Inclusionai | 552B | 12,559,070 |
| 40 | ▲6% | GPT Oss 120b | OpenAI | 524B | 142,947,445 |
Usage data: OpenRouter public rankings API, weekly window ending 2026-10-15. Refreshed weekly.
Usage rankings FAQ
Which AI model is used the most?
DeepSeek V4.1 Flash processed the most tokens of any AI model in the week ending 2026-10-15 — about 37 trillion through OpenRouter alone, ahead of Step 5 Preview (19.5T) and GLM-5.3 Flash (13T). Among frontier models, Claude Opus 5.5 leads with roughly 5.9 trillion tokens.
Why do cheap models top the usage rankings?
Usage measures total tokens, and cheap flash-class models run in high-volume pipelines — classification, extraction, autocomplete — that process orders of magnitude more text than chat. A model can top usage while ranking 30th on quality benchmarks; that is why this usage board sits next to the AMR quality leaderboard rather than replacing it.