62187385135356
Sim, perguntei ao Grok e seu comentário faz total sentido. Mas, ainda assim, é um indicativo do “mass market” de AI, digamos assim.
Trechos do que extrai do Grok:
OpenRouter acts as a unified API gateway (OpenAI-compatible) that lets you access 400+ models from 60+ providers (OpenAI, Anthropic, Google, DeepSeek, Qwen, Llama, etc.) through one key. It handles routing, fallbacks, and often cheaper/free options. No subscriptions—just pay-as-you-go credits with transparent per-model pricing (they sometimes add a small markup or take a cut on top of provider rates). Their site currently reports ~80T monthly tokens and 8M+ global users.
Scale and external validation
• Their big 2025 “State of AI” report analyzed over 100 trillion tokens of real usage (spanning ~2 years). It was a joint project with a16z (Andreessen Horowitz), who called it one of the most comprehensive empirical views of actual developer behavior available. a16z.com
• The paper is also on arXiv, and the data has been cited by analysts, news outlets, and researchers without accusations of faking numbers.
• Weekly volumes in the tens of trillions are consistent across months (e.g., similar Chinese-dominance weeks reported in March–April 2026 as well).
Mais especificamente sobre seu ponto:
OpenAI and Anthropic appear underrepresented on OpenRouter’s rankings because the platform naturally attracts price-driven traffic where cheap, high-volume Chinese models shine. This makes Chinese models look more dominant on this specific aggregator than they might in a full global picture (which includes direct premium usage). But the data itself isn’t skewed or unfair — it’s one of the best public signals of real developer behavior in the cost-sensitive segment. If you’re evaluating overall market leadership, you’d want to combine this with revenue figures, enterprise adoption reports, and benchmark performance, not rely on one aggregator’s token counts alone.
02:13