WHAT NEW IN X11.06 FRI

X 上的 AI 合集档案 · 23 个合集 · 70,791 条

jevIMAGE

Opus 5.5 + Jev 的逐区块交易 agent 实战

有人把 Opus 5.5 与 Jev 组合成每区块决策、单次低于 100ms 的 24/7 交易 agent,附 6 页研究论文和 Rust 代码库,已实盘跑 3 天。

Kiyoro
@0xKiyoro2026.09.28收录原推发布于 2026.09.27
Opus 5.5 + Jev 的逐区块交易 agent 实战
Kiyoro @0xKiyoro原推发布于 2026.09.27
在 X 上打开原推

A TRADING AGENT THAT MAKES A REAL DECISION ON EVERY SINGLE BLOCK. 24/7. UNDER 100 MS EACH. That's what Opus 5.5 + Jev looks like when someone actually wires it up. A 6-page research paper on how to find profitable strategies with it, from scratch. A complete codebase. In Rust. 3 days running live, and the builder calls the results incredible. Here's why this stack is different. A normal LLM takes 3 to 30 seconds to read an order book. By the time it answers, the book has moved 40 times. Jev doesn't write. It returns a calibrated decision before the next block lands. The split that makes it work: → code computes everything that's math: spread, imbalance, volatility, inventory → Jev answers the judgment calls: is this trending, is this flow toxic, is this setup worth it → 6 of those questions go out in ONE call, at the latency of one → code turns the probabilities into position size → hard risk limits sit on top, and the model can never override them And the cost flips the whole game. A decision on every block, around the clock, runs about $10–25 a month. The same loop on a frontier LLM burns $50–150 an hour. Most people are still pasting charts into a chatbot and waiting 10 seconds for an opinion. This stack decides before the next block lands.

这条是引用着另一条写的:「用 Jev 从零搭高频交易 bot:每区块一次买卖决策」

x.com/0xkiyoro/status/2104180789642424752采集于 2026.09.28 10:04