Jev 模型发布：主打决策而非对话，输出 token 免费

CyrilXBT @cyrilXBT原推发布于 2026.09.22

在 X 上打开原推

This is the tweet everyone building AI agents needs to actually sit with. Diogo Almeida co-invented ChatGPT. Then asked himself the question everyone else skipped: Why have superhuman chat models not led to AGI. Two years in stealth later, he shipped the answer: Jev. $42 per BILLION input tokens. Output tokens are literally free. Not cheap. Free. Because the architecture makes them too cheap to meter. Named after Jevons paradox on purpose. When something gets radically cheaper, you don't use less of it, you use insanely more of it. 20 to 200x faster. 40 to 400x cheaper. Not on writing. On deciding. That's the part people are going to miss reading this too fast. We spent two years building chat models that talk beautifully and still can't be trusted to make a bounded decision without burning frontier pricing to do it. Jev is the first model actually built for the other half of what an agent does. Almeida calls this the shortest path to an AI based economic revolution. I'd actually go further. This might be the first frontier model launch that isn't trying to win a chat leaderboard, and that's exactly why it matters. Follow @cyrilXBT, I'm tracking this one closely as it rolls out. https://x.com/kenonews/status/2101656436136661163/video/1

这条是引用着另一条写的：x.com/1373665408193036293/status/210151513079829…