Nace AI 演示 Drex Max：浏览器决策模型对比 Jev 1.14
Nace AI 展示面向 computer use 的多模态决策模型 Drex Max，称其单 token 成本更低、Mind2Web 准确率 88.3%，并与 Jev 1.14 的延迟和准确率做对比。
The real challenge with AI at scale is often the cost of making millions of decisions every day. @NaceAI recently shared an interesting demo featuring Drex Max, their multimodal decision model built for computer use and browser use. Think about how AI agents interact with a browser. Every task can involve dozens of small decisions, often requiring a new screenshot to be processed at each step. That's where inference economics becomes interesting. According to Nace AI, Drex Max offers: → Up to 167x lower per-token costs than models powering computer use from Claude, GPT, and Gemini → 88.3% accuracy on Mind2Web, compared to 86.5% for Jev 1.14 → 174 ms median decision latency, compared to 253 ms for Jev 1.14 in Nace AI's vision benchmark What stands out to me is how specialized models can make real-world AI agent deployments more economically viable while maintaining strong decision accuracy. When an agent needs to make dozens of decisions to complete a single task, even small improvements in latency and inference cost can add up. Of course, broader real-world testing matters, but these results highlight an important direction. AI adoption at scale depends on more than intelligence. It depends on the economics of every decision. Explore Drex Max: https://console.nace.ai #AI #BrowserAutomation #MultimodalAI #AIAgents Sharing the demo below 👇
这条是引用着另一条写的：x.com/naceai/status/2112175183448461683