LLM Rankings by Real-World Usage
Which large language models are developers actually shipping? Rankings are pulled live from OpenRouter, the largest neutral LLM gateway, and ranked by 7-day token throughput. This is a popularity board, not a quality board.
Last updated: 2026-08-16
Top 25 LLMs by 7-Day Token Usage
| # | Model | 7d Tokens | Apps | Context | $/M in | $/M out |
|---|---|---|---|---|---|---|
| 1 | Google: Gemini 3.1 Flash Lite Google | — | — | 1049k | $0.25 | $1.50 |
| 2 | Baidu Qianfan: CoBuddy (free) baidu | — | — | 131k | $0.00 | $0.00 |
| 3 | OpenAI: GPT Chat Latest OpenAI | — | — | 400k | $5.00 | $30.00 |
| 4 | SpaceXAI: Grok 4.3 xAI | — | — | 1000k | $1.25 | $2.50 |
| 5 | IBM: Granite 4.1 8B ibm-granite | — | — | 131k | $0.05 | $0.10 |
| 6 | Mistral: Mistral Medium 3.5 Mistral | — | — | 262k | $1.50 | $7.50 |
| 7 | Owl Alpha openrouter | — | — | 1049k | $0.00 | $0.00 |
| 8 | NVIDIA: Nemotron 3 Nano Omni (free) nvidia | — | — | 256k | $0.00 | $0.00 |
| 9 | Poolside: Laguna XS.2 (free) poolside | — | — | 262k | $0.00 | $0.00 |
| 10 | Poolside: Laguna M.1 (free) poolside | — | — | 262k | $0.00 | $0.00 |
| 11 | Anthropic Claude Haiku Latest ~anthropic | — | — | 200k | $1.00 | $5.00 |
| 12 | OpenAI GPT Mini Latest ~openai | — | — | 400k | $0.75 | $4.50 |
| 13 | Google Gemini Pro Latest ~google | — | — | 1049k | $2.00 | $12.00 |
| 14 | MoonshotAI Kimi Latest ~moonshotai | — | — | 1049k | $2.80 | $14.00 |
| 15 | Google Gemini Flash Latest ~google | — | — | 1049k | $0.38 | $1.88 |
| 16 | Anthropic Claude Sonnet Latest ~anthropic | — | — | 1000k | $2.00 | $10.00 |
| 17 | OpenAI GPT Latest ~openai | — | — | 1050k | $5.00 | $30.00 |
| 18 | Qwen: Qwen3.5 Plus 2026-04-20 qwen | — | — | 1000k | $0.30 | $1.80 |
| 19 | Qwen: Qwen3.6 Flash qwen | — | — | 1000k | $0.19 | $1.13 |
| 20 | Qwen: Qwen3.6 35B A3B qwen | — | — | 262k | $0.15 | $1.00 |
| 21 | Qwen: Qwen3.6 Max Preview qwen | — | — | 262k | $1.03 | $6.16 |
| 22 | Qwen: Qwen3.6 27B qwen | — | — | 262k | $0.30 | $2.00 |
| 23 | OpenAI: GPT-5.5 Pro OpenAI | — | — | 1050k | $30.00 | $180.00 |
| 24 | OpenAI: GPT-5.5 OpenAI | — | — | 1050k | $5.00 | $30.00 |
| 25 | DeepSeek: DeepSeek V4 Pro DeepSeek | — | — | 1049k | $1.17 | $2.34 |
Why Usage Rankings Matter More Than Benchmarks
Every quarter brings a new "state-of-the-art" LLM that tops some benchmark. Most of them never see real production traffic. Usage rankings cut through the noise — they show which models developers actually trust enough to put in front of paying customers.
OpenRouter is the right place to measure this because it's provider-neutral. Developers route through OpenRouter when they want flexibility — A/B testing models, falling back across providers, optimizing for cost or latency. The traffic mix on OpenRouter is the closest thing to an honest market signal for which LLMs work in real apps.
For a more rounded view that includes coding benchmarks and human preference scores, see the full LLM Leaderboard. For curated picks for specific use cases, check our Best LLM Models guide.
Frequently Asked Questions
What is OpenRouter and why use its data?
OpenRouter is a unified API gateway for 200+ LLMs across providers like Anthropic, OpenAI, Google, Meta, DeepSeek, and xAI. Because it's provider-agnostic, the traffic mix is the cleanest signal we have for which models the market actually prefers.
How is "7-day tokens" calculated?
Total input + output tokens routed through OpenRouter to that model over the last 7 days, summed across all apps and developers. We pull this directly from OpenRouter's public rankings page.
Does usage rank correlate with quality?
Loosely. The top 5 are usually high-quality models that are also affordable enough to ship. But cheap models with mediocre quality can rank high if they're used for high-volume tasks like embeddings or classification. Cross-reference with the Arena and coding benchmarks before picking a default.