Models are the supply side; apps are the demand. This maps the ~43 ranked apps to the models and vendors they run on — revealing how concentrated each vendor's volume is on a handful of agent apps.
Where the demand actually goes: each top app's tokens flowing to the vendors it runs on, then to the model's country of origin. Ribbon width = tokens; red = Chinese-origin, blue = Western.
The same demand, resolved to the concrete models each app routes to (top 9 models; the rest grouped as "Other models").
| app | tokens | share | runs on (vendor mix) |
| Hermes Agent | 45.2T | 25.1% | deepseek 51% · z-ai 10% · stepfun 8% |
| Claude Code | 43.9T | 24.4% | anthropic 46% · deepseek 26% · z-ai 19% |
| Kilo Code | 22.4T | 12.4% | nvidia 46% · minimax 15% · stepfun 10% |
| pi | 16.2T | 9.0% | deepseek 62% · z-ai 14% · anthropic 7% |
| OpenClaw | 12.3T | 6.8% | stepfun 54% · deepseek 21% · minimax 10% |
| Codex | 11.9T | 6.6% | openai 69% · deepseek 21% · z-ai 4% |
| Cline | 7.6T | 4.2% | xiaomi 28% · meta 15% · stepfun 14% |
| Portkey AI | 5.0T | 2.8% | z-ai 62% · typesafe 18% · deepseek 11% |
| OpenHands | 2.8T | 1.5% | deepseek 74% · z-ai 19% · moonshotai 3% |
| ISEKAI ZERO | 2.3T | 1.3% | xiaomi 49% · deepseek 22% · z-ai 19% |
| Lemonade | 1.5T | 0.8% | z-ai 77% · openai 9% · inclusionai 7% |
| LangChain | 1.3T | 0.7% | deepseek 55% · openai 13% · z-ai 9% |
| Descript | 1.2T | 0.7% | anthropic 67% · google 18% · openai 13% |
| Nous Research API | 1.1T | 0.6% | deepseek 45% · poolside 14% · stepfun 9% |
| Zed Editor | 889.5B | 0.5% | deepseek 50% · z-ai 18% · openai 9% |
| Janitor AI | 764.4B | 0.4% | deepseek 55% · nvidia 19% · z-ai 18% |
| Craft | 571.2B | 0.3% | z-ai 35% · openai 26% · meta 13% |
| GDevelop | 557.0B | 0.3% | meta 60% · openai 18% · deepseek 12% |
| Mira | 524.4B | 0.3% | minimax 83% · qwen 5% · openai 4% |
| Open WebUI | 494.0B | 0.3% | deepseek 45% · z-ai 17% · openai 11% |
| SillyTavern | 334.7B | 0.2% | deepseek 34% · z-ai 23% · google 22% |
| extra.email | 222.2B | 0.1% | z-ai 61% · typesafe 35% · deepseek 3% |
| Roo Code | 192.1B | 0.1% | deepseek 60% · z-ai 10% · openai 7% |
| nanobot | 173.8B | 0.1% | deepseek 69% · xiaomi 9% · openai 8% |
| Qwen Code | 116.7B | 0.1% | deepseek 51% · qwen 22% · z-ai 9% |
Share = % of tokens across all tracked apps. "Runs on" = the app's top vendors by token volume.
| vendor | tokens (via apps) | top-app conc. | top driving apps |
| deepseek | 57.4T | 40% | Hermes Agent 40% · Claude Code 20% · pi 18% |
| anthropic | 25.8T | 77% | Claude Code 77% · Hermes Agent 10% · pi 5% |
| z-ai | 24.1T | 35% | Claude Code 35% · Hermes Agent 19% · Portkey AI 13% |
| nvidia | 13.9T | 75% | Kilo Code 75% · Hermes Agent 13% · Cline 6% |
| stepfun | 13.6T | 48% | OpenClaw 48% · Hermes Agent 26% · Kilo Code 16% |
| openai | 12.4T | 66% | Codex 66% · Hermes Agent 14% · pi 8% |
| minimax | 10.0T | 34% | Kilo Code 34% · Hermes Agent 31% · OpenClaw 12% |
| xiaomi | 5.2T | 40% | Cline 40% · ISEKAI ZERO 22% · Hermes Agent 21% |
| upstage | 4.1T | 73% | Hermes Agent 73% · Cline 24% · Nous Research API 2% |
| poolside | 2.6T | 57% | Kilo Code 57% · Hermes Agent 31% · Cline 6% |
Top-app conc. = share of the vendor's app-driven volume from its single biggest app — high (≥40%) = concentration risk, low (<20%) = broad, durable demand.