在 OpenRouter 上——一个供开发者将 AI 模型接入其产品的平台——自 2025 年 1 月以来，每周令牌消耗量激增了超过 25,000%，从 0.5 万亿增加到 126.2 万亿令牌。令牌是 AI 处理的基本单位，就像汽车使用的加仑汽油一样。但这张图表反映的，与其说是 AI 的广泛应用，不如说是令牌指标的膨胀。
这种激增不应被误认为实际使用量或商业价值的相应跃升：https://the-decoder.com/ai-bubble-debate-google-prefers-to-mention-token-consumption-rather-than-sales-figures/。推理模型在产生答案前会生成大量“思考”令牌，极大地抬高了总数。使用量的轻微增加就可能意味着令牌消耗的巨大激增：https://the-decoder.com/frontier-radar-3-how-agentic-ai-is-turning-tokens-into-a-business-metric/，尤其是在未经优化的具代理性 AI 系统中，它们以惊人的速度消耗令牌。
On OpenRouter, a platform developers use to plug AI models into their products, weekly token consumption has surged more than 25,000 percent since January 2025, from 0.5 trillion to 126.2 trillion tokens. Tokens are the basic unit of AI processing, like gallons of gas for a car. But this chart says less about booming AI adoption than about how inflated token metrics have become.
The surge shouldn't be confused with a matching jump in actual usage or business value：https://the-decoder.com/ai-bubble-debate-google-prefers-to-mention-token-consumption-rather-than-sales-figures/. Reasoning models generate massive amounts of "thinking" tokens before producing an answer, inflating the count dramatically. A small uptick in usage can mean a huge spike in token consumption：https://the-decoder.com/frontier-radar-3-how-agentic-ai-is-turning-tokens-into-a-business-metric/, especially from unoptimized agentic AI systems that burn through tokens at staggering rates.
OpenAI's GPT 5.6 Luna recently dominated token consumption on OpenRouter：https://x.com/OpenRouter/status/2099898254905549220, though that doesn't necessarily mean more people are using it. The model may simply generate more tokens per prompt. On the revenue side, OpenAI's Astra leads. Chinese models like Kimi, GLM, and DeepSeek are growing fast too, with monthly spending up tenfold in 2026, though from a much smaller base. Ad Ad
Stay in the loop on AI. Clear, useful, no fluff.
Follow The Decoder for AI news, background stories and expert analyses.
扫码直达文章详情