Drex Max 决策模型延迟 183ms，比 Claude Opus 5.5 快约 19 倍
NaceAI 的 Drex Max 不生成文字，直接吐概率，183ms 完成一次决策，比 Claude Opus 5.5 的 3.5 秒快近 20 倍。做需要大量高频决策的智能体的开发者可以关注，延迟敏感场景里这差距是实打实的。
NaceAI 发布新论文，介绍 256k 上下文的决策模型 Drex Max。实测延迟仅 183 毫秒，而 Claude Opus 5.5 为 3.5 秒，GPT-6 Sol 约 2 秒。
技术上，Drex Max 完全跳过文本生成：输入一组选项加上文本、截图或图表，单次前向传播直接输出校准后的概率。Artificial Analysis 的独立 Decision Index 排行中，Drex Max 1.1 版位列决策模型类第一名。
对开发者而言，如果一个智能体任务需要做数百次小决策，每次省下 2-3 秒，累积耗时差异会非常显著。这类专用决策模型与通用推理模型形成了明确的分工路线。
183 milliseconds vs. 3.5 seconds for Claude Opus 5.5.
That's the latency gap @NaceAI reports in its new paper on Drex Max, a 256k context decision model. GPT-6 Sol comes in around 2 seconds.
Drex Max skips text generation entirely. You hand it a set of options, plus text, screenshots or charts, and it returns calibrated probabilities in one pass. @ArtificialAnlys just ranked version 1.1 #1 among decision models on its independent Decision Index.
If you build agents that make hundreds of small choices per task, those seconds add up fast.
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Drex Max, Claude, GPT-6, 智能体.
Source attribution: curated from kimmonismus (@kimmonismus). Cross-source corroboration: 5 independent sources have covered this development (including Aran Komatsuzaki (论文推介), elvis, Artificial Analysis), confirming industry-wide signal strength.
