Three questions: §1 is AI token demand growing? (TAM) §2 which models are winning / fading / new? (popularity) §3 what are those tokens being spent on? (direction — coding? agent? chat? multimodal?).
Source: daily scrape of openrouter.ai/rankings. OpenRouter ≠ whole market (direct-API enterprise invisible), but it's the only public window into cross-provider inference mix + the long-tail dev-tool segment.
Absolute inference demand. Is the pie expanding?
Which models are winning this week, which are fading, which are new.
| # | Model | Provider | Tokens | Share | Requests | Trend | Velocity | Variants |
|---|---|---|---|---|---|---|---|---|
| 1 | ●tencent/hy3-preview | tencent | 3.75T | 14.8% | 71,017,013 | → +0% | free,standard | |
| 2 | ●moonshotai/kimi-k2.6 | moonshotai | 1.72T | 6.8% | 29,961,712 | → -3% | aggregate,standard | |
| 3 | ●anthropic/claude-4.6-sonnet | anthropic | 1.43T | 5.7% | 38,030,441 | → +4% | aggregate,standard | |
| 4 | ●anthropic/claude-4.7-opus | anthropic | 1.14T | 4.5% | 19,401,460 | → +8% | aggregate,standard | |
| 5 | ●google/gemini-3-flash-preview | 1.06T | 4.2% | 135,590,969 | → +3% | aggregate,standard | ||
| 6 | ●deepseek/deepseek-v4-flash | deepseek | 976.0B | 3.9% | 64,838,418 | → +6% | aggregate,standard | |
| 7 | ●deepseek/deepseek-v3.2 | deepseek | 837.4B | 3.3% | 101,031,820 | → +4% | aggregate,standard | |
| 8 | ●minimax/minimax-m2.7 | minimax | 766.8B | 3.0% | 26,021,878 | → +1% | aggregate,standard | |
| 9 | ●deepseek/deepseek-v4-pro | deepseek | 763.7B | 3.0% | 17,596,540 | → +7% | aggregate,standard | |
| 10 | ●x-ai/grok-4.1-fast | x-ai | 730.7B | 2.9% | 117,258,584 | → +1% | aggregate,standard | |
| 11 | ●google/gemini-2.5-flash-lite | 656.1B | 2.6% | 250,546,747 | → +0% | aggregate,standard | ||
| 12 | ●google/gemini-2.5-flash | 587.0B | 2.3% | 159,267,019 | → +0% | aggregate,standard | ||
| 13 | ●stepfun/step-3.5-flash | stepfun | 559.9B | 2.2% | 9,336,499 | → -4% | aggregate,standard | |
| 14 | ●anthropic/claude-4.6-opus | anthropic | 546.3B | 2.2% | 12,115,480 | 🚀 +20% DoD | aggregate,standard | |
| 15 | ●nvidia/nemotron-3-super-120b-a12b | nvidia | 532.5B | 2.1% | 14,786,343 | → +1% | free,standard | |
| 16 | ●openai/gpt-oss-120b | openai | 528.7B | 2.1% | 95,487,970 | → -0% | aggregate,free,standard | |
| 17 | ●z-ai/glm-5.1 | z-ai | 355.2B | 1.4% | 12,311,766 | → +2% | aggregate,standard | |
| 18 | ●google/gemini-3.1-flash-lite-preview | 330.6B | 1.3% | 68,836,281 | → +1% | aggregate,standard | ||
| 19 | ●openai/gpt-5.5 | openai | 322.9B | 1.3% | 7,489,946 | → +7% | aggregate,standard | |
| 20 | ●openrouter/owl-alpha | openrouter | 306.2B | 1.2% | 15,337,863 | 🚀 +28% DoD | aggregate,standard | |
| 21 | ●google/gemini-3.1-pro-preview | 284.9B | 1.1% | 14,461,048 | → -3% | aggregate,standard | ||
| 22 | ●moonshotai/kimi-k2.5-0127 | moonshotai | 270.3B | 1.1% | 13,667,445 | → +0% | aggregate,standard | |
| 23 | ●minimax/minimax-m2.5 | minimax | 251.3B | 1.0% | 12,748,430 | → +15% | aggregate,free,standard | |
| 24 | ●openai/gpt-4o-mini | openai | 242.3B | 1.0% | 117,043,023 | → +0% | aggregate,standard | |
| 25 | ●anthropic/claude-4.5-haiku | anthropic | 237.9B | 0.9% | 25,857,068 | → +2% | aggregate,standard | |
| 26 | ●openai/gpt-5.4 | openai | 236.6B | 0.9% | 13,127,779 | → +4% | aggregate,standard | |
| 27 | ●google/gemma-4-31b-it | 207.5B | 0.8% | 35,955,884 | → +4% | aggregate,free,standard | ||
| 28 | ●z-ai/glm-5 | z-ai | 207.2B | 0.8% | 10,216,802 | 🚀 +25% DoD | aggregate,standard | |
| 29 | ●qwen/qwen3.6-plus-04-02 | qwen | 193.3B | 0.8% | 6,340,635 | → -3% | aggregate,standard | |
| 30 | ●poolside/laguna-m.1 | poolside | 185.2B | 0.7% | 3,965,170 | → +12% | free,standard |
What are the tokens being used FOR — token types (thinking / cached / tool / multimodal) + app categories (coding / agent / chat / ...)
| # | App | Category | Tokens | Requests | Origin |
|---|---|---|---|---|---|
| 1 | OpenClaw OpenClaw is an open-source AI agent that connects to your messaging apps and takes real actions on your behalf, from run | ⚡ Productivity | 1.83T | 34,318,499 | https://openclaw.ai/ |
| 2 | Hermes Agent Hermes Agent is an open-source, self-improving AI agent by Nous Research that runs persistently with memory across sessi | 🤖 General agent | 1.68T | 34,664,853 | https://hermes-agent.nousresearch.co |
| 3 | Kilo Code Kilo Code is an open-source AI coding agent that works across VS Code, JetBrains, and CLI to help developers ship code f | 💻 Coding agent | 1.22T | 18,244,426 | https://kilocode.ai/ |
| 6 | Claude Code Claude Code is Anthropic's agentic coding tool that reads your entire codebase, plans and executes changes across files, | 💻 Coding agent | 633.3B | 8,612,852 | https://claude.ai/ |
| 7 | Descript AI Video & Podcast Editor | 💻 Coding agent | 316.6B | 5,963,832 | https://descript.com/ |
| 8 | pi There are many coding agents, but this one is yours. | 💻 Coding agent | 306.1B | 4,651,115 | https://pi.dev/ |
| 13 | Janitor AI Janitor AI is a chatbot platform where users create and chat with custom AI characters for interactive roleplay, storyte | 💬 Chat / roleplay | 192.1B | 14,989,759 | https://janitorai.com/ |
| 14 | ISEKAI ZERO AI adventures. Travel with your favorite characters | 💬 Chat / roleplay | 184.6B | 7,646,345 | https://www.isekai.world/ |
| 15 | Cline Cline is an open-source AI coding agent that lives inside your IDE, autonomously exploring your codebase, editing files, | 💻 Coding agent | 167.4B | 1,971,857 | https://cline.bot/ |
| 16 | Lemonade The AI tool for Roblox games. | 🎮 Gaming | 152.2B | 4,202,828 | https://lemonade.gg/ |
| 17 | Roo Code Roo Code is a VS Code extension that gives you a team of specialized AI agents with customizable modes for coding, archi | 💻 Coding agent | 151.8B | 1,822,680 | https://github.com/RooVetGit/Roo-Cli |
| 18 | VidMuse Generate music videos, shorts, and more from audio | 💻 Coding agent | 60.2B | 1,458,513 | https://vidmuse.ai/ |
| 19 | SillyTavern SillyTavern is the LLM frontend for power users, a chat interface that connects to any model and gives deep control thro | 💬 Chat / roleplay | 55.7B | 3,303,759 | https://sillytavern.app/ |
| 20 | LangChain | 🏗 Framework / infra | 49.6B | 3,905,143 | https://docs.langchain.com/ |
| 24 | Mira is the leading AI agent inside Telegram Mira is a Telegram-native AI assistant. | 💻 Coding agent | 43.8B | 2,823,682 | https://mira.tg/ |
| 25 | MinutaIA Brazilian AI platform for law | 🏗 Framework / infra | 39.6B | 1,394,040 | https://minutaia.com.br/ |
| 26 | HammerAI Chat with AI characters for free | 💬 Chat / roleplay | 39.0B | 11,357,506 | https://www.hammerai.com/ |
| 29 | wsup.ai - Chat with AI Characters Online for Free | 💬 Chat / roleplay | 35.8B | 16,618,366 | https://wsup.ai/ |
| 30 | Chub AI GenAI for everyone | 💬 Chat / roleplay | 32.5B | 2,446,555 | https://chub.ai/ |
| 31 | Zed Editor AI code editor designed for high-performance collaboration | 📦 Other | 31.3B | 448,714 | https://zed.dev/ |
Source: https://openrouter.ai/rankings (public, no auth). Daily Flight-payload scrape via openrouter_model_share.py.
Data files: daily_snapshots.jsonl (388 model-days on first run) · daily_apps.jsonl (60 app-ranks × 3 buckets).
Caveats: (i) OpenRouter skews to long-tail + dev tools — direct API traffic is invisible. (ii) App categorization: Keyword-based categorization (brittle). Feed back misclassifications to APP_CATEGORIES dict. (iii) Token-type metrics depend on providers reporting the underlying fields (some omit reasoning/cached); the "0%" rows are likely under-reporting, not zero.
Generated 2026-05-09 05:45 UTC