어제 나온 Jev, 하루 만에 실사용 후기가 쌓였다 — "LLM이 아니라 판단 엔진"
ChatGPT 공동 개발자가 어제 공개한 Jev를 실제로 돌려본 이야기가 하루 만에 여럿 나왔다. 한 개발자는 분류·모델 라우팅·의도 파악 등에 5,000번 요청을 보내고 2달러가 들었다고 했다. 텍스트를 생성하는 대신 질문에 대한 판단 결과와 확률을 한 번에 돌려주는 구조라, 코드로 짠 규칙과 LLM 호출 사이의 새로운 선택지라는 평가다. 입력 요금이 GPT-5.6 Sol의 약 95분의 1이고 출력은 무료라 LLM-as-a-judge 같은 용도에 맞는다는 얘기가 붙었다. 다만 생성 모델이 아니라서 대부분의 작업에서 LLM을 대체하지는 못한다는 지적도 함께 나왔다.
출처 6건 보기· @MichaelLee04, @polymarketjapan, @Yuchenj_UW 외 3
- @MichaelLee04I got access to Jev earlier today (thank you @hackgoofer). I have run ~5,000 requests so far, (which cost me around $2!), across classification, model routing, intent, steering, and many other things. tl;dr, Jev enables a new intelligent decision-making primitive, separate from deterministic code and LLM calls. This allows a class of decision-making that was neither suited to dumb, unintelligent code, nor to slow, expensive LLMs. It is super fast and cheap, and I think I will likely end up mX ♥2.9천
- @polymarketjapan【速報】元OpenAI研究者、新型AI「Jev」を発表 ・文章生成ではなく、質問への判断結果と確率を一括で返すソフトウェア向けAI ・GPT-5.6 Sol比で入力料金は約95分の1、出力料金は無料 ・特定用途では通常のLLMより20~200倍高速と主張X ♥2.1천
- @Yuchenj_UWJev has spoken. It picked which model is AGI. 20–200x faster. 40–400x cheaper. This could make things like LLM-as-a-judge insanely fast and nearly free. (I tried a bunch of prompts and still didn’t burn through $0.10.)X ♥1.9천
- @kunchenguidholy crap this is super exciting it’s not “generative”, so imo it’s a bit misleading to compare it with LLM. it can’t replace LLMs on most tasks but it can automate a lot of things where the options are clear but choice is not - it’s basically a real time, almost free, intelligent decision engine i can already think of so many practical use cases. can’t wait to get my hands on it!X ♥774
- @emollickThis is an excellent demo of an AI model designed to rapidly classify information according to decision rules, much faster and cheaper than an LLM. It is not an LLM itself and cannot output text or code, but if it works, has a useful role in systems. (Haven’t tried it yet myself)X ♥433
- @NielsRoggeFor anyone curious how Jev works, I made a visual explanation using @claudeai :) This is based on the Qwen2.5-RLCD model which @harshagundal released on @huggingface The idea is to replace autoregressive LLM generation by a single Transformer decoder (of a pre-trained LLM), which processes the context + JSON schema only once. The keys and values of those tokens are cached. Next, for each field of the JSON schema, we: 1. pass its field suffix tokens through the Transformer decoder again (reusX ♥1.4천