ego lite + Jev + DeepSeek Flash = stupid fast. ⚡ 3.71s for 20 Amazon product decisions. GPT-5.6 Sol: 54.45s Same task. Same result. 10/10 on both. Less reasoning. Faster decisions. 1× speed.

就知道你会结合 jev，太合适了

内存泄漏好像有点严重，一下子4、5g的内存占用

Thanks for flagging this. Could you share a bit more about what was happening when memory usage reached 4–5GB? For example, how many Spaces/tabs were open, whether an agent was running, and whether the memory dropped after closing the Space. We’ll dig into it on our side too.

就一个，我当时多开了4、5个标签页就开始莫名的卡了，而且没有代理在运行

Speed is great, but I worry about edge cases.

这个工作流具体如何实现呢？

怎么把jev加入egoblite的工作流呢，什么时候出官方支持

when is on linux? pliz

我也是这么组合的 不过是 ego lite +jev +6astra 不过速度慢

same 10/10 in 3.71s vs 54s means Sol spent about 50s you did not need. were the 20 Amazon calls one Jev batch, or 20 sequential hits? that is what I would copy

How to use deepseak with jev

那怎么结合使用呢

Both at 10/10 is the number that matters: 50 extra seconds bought zero accuracy, so you're paying for insurance, not thought. Benchmarks grade the destination and ignore the fuel bill. This chart is the easy half; the honest test is where Sol's extra minute earns its keep.

我什么时候能用上

How to??

How do we get to try this?

@grok 这个工作流是怎么实现的？你能详细分解说明一下吗？

Linux Linux Linux

wild speed win. the quiet failure mode on "same result 10/10": agreement ≠ calibrated confidence. if you only score accuracy on the easy batch, near-tied low-conf Choices still look like wins until a high-stakes SKU flips. keep confidence as a second axis and shadow the auto-act path until the conf/acc curve looks honest on hard cases. also: batch independent Choices in one System One call — that's where a lot of the wall-clock comes from.

for my computer use/ browser use i am having issues. Any skill for it?

That speed gap is seriously impressive

When Windows version？

You know what I've been thinking lately 😍, this is the speed at which I want the results.

same result but way faster, thats the whole pitch