I totally can't sleep now....

Matched 15 brands to real-time news hooks and applied intent labels across 428 items in 28 seconds, comparing Jev against DeepSeek.

The original post, translated

I totally can't sleep now.... This video lets you completely understand JEV's terrifying capability

I actually demonstrated it, using JEV and DeepSeek to do the same task: from the massive real-time news stream of the day, quickly match for 15 brands whether there are suitable PR hot topics to piggyback on, and tag them with structured intent classification labels

The result is: within 28 seconds, Jev blasted through a full 428 pieces of data, and completed all the goal decomposition and classification; while at the same time, DeepSeek V4.1-flash running on the same task had only just completed 6/428 items

The speed gap was widened by dozens of times, and the answer even returned before you finished reading the first line....

Why is there such a terrifying gap? The underlying logic is:

• General models like DeepSeek are essentially "writing an article word by word":

even if you only ask it to make the simplest "yes/no" judgment, in the background it still has to push out one word at a time, forcibly going through a long generation process, with latency starting at seconds

• Jev is officially positioned as a "System 1 (fast decision)" model, and fundamentally it doesn't write at all:

it completely abandons word-by-word output, and at the underlying level directly produces results in parallel all at once. It doesn't output half a sentence of nonsense, and within 70 to 200 milliseconds directly throws out the final answer the code needs (right/wrong, score, classification)

demo reference @elvissun Github open source:

I suggest everyone try JEV, the more you use it the more awesome it feels.....

The post above is a machine translation from zh; the untranslated text is in the fold-out.

Quoted post

S

@SUOHA_AI45.4K views

source

很多人没看懂爆火的JEV是什么？ 所以我做了一个 50 秒的演示视频，带大家把海外刷屏的 JEV 以及它背后的底层架构看透👇（见下方视频） JEV 是前 OpenAI、参与 ChatGPT / RLHF 的 @CompleteSkeptic（Diogo Almeida）创办 @typesafeai 后的第一款模型 他在 OpenAI 时发现了一个致命矛盾：

What the author linked

• github.com/elvisun/newsja…
• GitHub - elvisun/newsjack: The open-source skills that turn your agent into a full PR team. — The open-source skills that turn your agent into a full PR team. - elvisun/newsjack

Engagement when collected

Numbers are a snapshot taken from X when the case was added to the library (schema v1, collected 2026-09-19); they will not match today.