By real usage — the total token volume routed per model on OpenRouter over the last 7 days (updated 2026-07-17) — the most-used AI models right now are Hy3 (free) (tencent, 9.79T), MiMo-V2.5 (Xiaomi, 8.68T), DeepSeek V4 Flash (DeepSeek, 5.24T). The surprising part: 7 of the top 10 are open-weight models, and Chinese open models (from labs like DeepSeek, Xiaomi, Tencent, MiniMax and Z.ai) dominate the top of the list by token volume. OpenAI ranks lower than most people expect on this measure: the first OpenAI model, GPT-5.5, sits at rank 14. This tracks OpenRouter API/marketplace demand, so first-party ChatGPT traffic is not counted.
| # | Model | Tokens/wk |
|---|---|---|
| 1 | Hy3 (free) |
9.79T
|
| 2 | MiMo-V2.5 |
8.68T
|
| 3 | DeepSeek V4 Flash |
5.24T
|
| 4 | MiniMax M3 |
3.95T
|
| 5 | GLM 5.2 |
3.52T
|
| 6 | Nemotron 3 Ultra (free) |
3.32T
|
| 7 | DeepSeek V4 Pro |
2.47T
|
| 8 | Claude Opus 4.8 |
2.2T
|
| 9 | Claude Opus 4.7 |
2.11T
|
| 10 | Gemini 3 Flash Preview |
947B
|
| 11 | Claude Sonnet 4.6 |
942B
|
| 12 | Claude Sonnet 5 |
937B
|
| 13 | Step 3.7 Flash |
894B
|
| 14 | GPT-5.5 |
678B
|
| 15 | MiMo-V2.5-Pro |
634B
|
Ranked by real usage: the total tokens (prompt + completion) routed to each model on OpenRouter over the trailing 7 days. This reflects developer and API demand on the OpenRouter marketplace and does not include first-party app traffic (ChatGPT, the Gemini app, claude.ai), so consumer chat flagships are under-counted relative to their true reach. Names, pricing and context windows are live from OpenRouter's own catalog and validated against it before display. Auto-refreshed; no hand-maintained list.