Jev just landed and the agent stack moved in three days. TypeSafe AI released Jev on September 15. It is built for the work agents actually do most of the time: choose the next tool, score a risk, decide whether to retry, or answer yes or no. Latency is under half a second. Show more
CommunityPractices & Patterns
Jev's Architecture Unmasked
A roughly 10,000-call black-box investigation that reconstructs Jev's likely architecture: a shared state prefix, isolated question branches, and direct probability readouts.
- Category
- Practices & Patterns
- Format
- Article
- Published by
- Community
- Author
- archerhume
- Added
- 2026-09-25
- Last verified
- 2026-09-25
Highlights
- Argues output_tokens is a billing artifact: yes/no counts fit four shared tokens plus 15 per answer.
- Token accounting is additive: one yes/no question used 268 input tokens and two used 276.
- A 23k-token state plus 5,000 questions fits the roughly 65,536-token cap implied by the counts.
- Latency barely moves up to about 100 questions, and 1,500 questions returned with a 610 ms median.
- On 1,200 MMLU items it measured ECE 0.031, and an option reversal shifted probability from 0.84-0.89 to 0.93-0.96.
Watch out
Self-published black-box inference with no vendor confirmation; the post labels KV sharing, readout type, and sparse MoE as progressively more speculative, and it targets jev-1.13.0 specifically.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: How the agent stack moved in Jev's first three days
YouTube: Diogo Almeida tells swyx what System One is, why RLHF failed his automation goal, and why AI should eventually disappear.
X: An A/B test against 4,000 personas
Jev is WILD I gave it two launch tweets and fed it over 4000 demographic profiles of real survey participants Twelve seconds later, a simulated A/B test tied to actual personas voting on the best tweet you can just do things i made it 100% free (link below)
X: Ad creatives from filtered assets
Jev is really good at content filtering and asset selection. DeepSeek + Higgsfield turn the selected assets into ad creatives.
More like this
A critical technical read of Jev's launch claims that digs into the calibration and eval numbers and asks what the architecture keeps secret.
Practices & PatternsArticle#community#evaluation#calibration
Community
Notes from a technical analysis that inferred Jev's internals from roughly 1,000 API calls, probing how probabilities and confidence behave.
Practices & PatternsPost#community#x#analysis
Community
A comparative review of Jev and the open-source Laya encoder: interfaces, deployment, disclosed internals, calibration claims, and the limits of agreement-based workflow evals.
Practices & PatternsArticle#community#evaluation#calibration
Community
From the community
Posts from builders shipping with Jev right now.
Reranking 33,047 catalog entries
拿 Jev 做搜索重排，我先泼一盆冷水：单独用，它没打赢向量检索 TypeSafe 的 Jev 这阵子很火，一堆项目拿它做重排。我们在 Agent Skills Hub 的 33,047 条目录上认真测了一次，164 条中英文真实查询，9,831 对分级标注，整套只花了 2.6 美元 三个结论 01｜单独重排，约等于没赢 Jev 重排 bge-m3 Show more
有美团、阿里的老哥嘛？ 试试加一路召回、重排（离线、近实时实现），我觉得有奇效 他在文本理解上 跟之前机器学习、llm很不一样 还能自动打标签做特征
Six uses that stuck after 60 days
I've been using Jev by @typesafeai Here's the six things i've tried and am confident I'll still use Jev for 60 days from now. There's many more experiments, ideas, and things I think I will use it for. It's a big deal (more on why in next post). But I am only sharing things Show more
Security decisions that fit Jev
This made me rethink where AI actually fits into security engineering. For purely engineering work, forget about ChatGPT or Claude. TypeSafe AI just released Jev, and I think it’s going to change how we build AI into security workflows. Instead of asking an LLM to “investigate Show more
we are officially out of stealth! join the frontier and get access to Jev on our website (link on profile)
A million judged questions
Ask Jev anything. Give it a try at askjev.ai It won't answer. It will judge. Let's see if we can get to 1 million questions. @typesafeai 🤝 @convex work great together. @hmartenjoyer @CompleteSkeptic @justKDeng @mikeysee
Inferring Jev's internals from 1,000 calls
Jevの内部アーキテクチャを推測している技術記事(Jev’s Architecture Unmasked)からメモ。 ・本記事はJevのAPIを約1万回の呼び出して、内部構造を推測したもの ・従来の言語モデルを用いた分類やルーティングでは、トークンを1文字ずつ逐次生成するために膨大な無駄な計算コストが発生していた。 Show more
The open System One roundup
Jev 发布没几天，开源社区已经开始疯狂复刻了🔥 最值得推荐的五个模型： 1、Laya 421M：原生决策模型，支持 Mac 2、Decider-2B：最像 Jev，基于 Qwen3.5 3、NanoJev 0.6B：专门的 Decision Head 4、Reflex：Qwen3.5 + Direct Logits 5、System-One 4B：专门做概率校准 Show more
Jev 刚发布没几天，开源社区就出现了同款🔥 Decider-2B模型，是基于 Qwen3.5-2B 做了特殊调整 它和 Jev 模型是一样的 只做选择 评分和判断 不是文本类的 LLM 模型 但两者还是有几个明显区别： 1、模型 Jev：闭源 System One Model Decider：Qwen3.5-2B，约 1.9B 参数，Apache 2.0 开源 2、价格
