TypeSafe introduces Jev, an AI model for decisions inside software

@omarsar0Recommended read. Jev gives up text generation to make AI dramatically faster. TypeSafe built a new architecture that answers structured questions in parallel, with RLCD training its probabilities to reflect how often it’s right. The team reports 40–200x faster responses on System One queries.

@rohanpaul_aiAnother brilliant launch for developers: and its 20-200x faster than LLMs because it skips token-by-token generation entirely. TypeSafe AI just launched Jev, > 20-200x faster >40-400x cheaper (w/ output tokens free) > Frontier composable intelligence optimized for decisions So Jev is an AI model built to make software decisions instead of writing text for people. A normal LLM answers by generating tokens sequentially, so software often has to request structured output, parse it, validate it, and decide what happens next. Jev removes that translation layer: give it some data and a predefined question, and it returns a typed choice or score with probabilities and confidence. For example, a support app can ask whether a ticket is urgent, whether it violates policy, and which queue should receive it, then act on those answers directly. In code, Jev behaves like a smart if-statement: ordinary software controls the workflow while the model handles fuzzy judgments that rigid rules struggle with. Jev evaluates multiple structured questions in parallel rather than writing an answer token by token. TypeSafe reports 70-500ms responses and 40-200x faster performance than comparable LLMs

@DCVCToday’s AI models require constant human supervision because they aren't optimized for autonomous execution. Backed by a $40M Series Seed led by @DCVC, @typesafeai is emerging from stealth with its first model, Jev—which delivers frontier intelligence at <100ms latency, running up to 100x faster and cheaper than existing frontier models. 💪 🔥 Built by @OpenAI veteran Diogo Almeida (@CompleteSkeptic), TypeSafe is turning AI into a machine-native utility. Great @Forbes coverage in 🧵

@michaelnemtsevAI Field Notes #106 Anthropic cut Claude Code's weekly limit 17% and called it a 25% raise. Within two days developers were routing the tool to cheaper rivals. - Google shipped Gemini 3.8 Live: voice models that switch 97 languages mid-call, live in the API - Shanghai's lab dropped Atria Dawn, an open 744B agentic model, MIT-licensed, no press release - TypeSafe's Jev claims it can't hallucinate, at $42 per billion tokens - An AI agent found admin access to Baseten's production code in 25 minutes - OpenAI, Anthropic and Google admit they've been coordinating on safety TODAY'S STORIES: 1. Claude Code limits: Anthropic trims weekly usage 17% and calls it a 25% raise If you run Claude Code all day, your weekly ceiling just dropped even though the email said otherwise. Check your real usage against the new numbers before mid-week, when heavy users hit the wall. Routing the tool to a cheaper model is now a setting, not a hack. 2. Gemini 3.8 Live: Google ships voice models that keep talking while they work Building a voice app or a support line? The bar for sounding human and getting something done just rose, and it runs behind an API you can call now. For a freelance developer, the gap between a demo and a shipped product shrinks, and clients will expect more for the same fee. 3. Atria Dawn: Shanghai lab drops a 744B open agentic model with no press release A capable agent you can download and run in-house changes the math for any engineering team wary of sending code or customer data to a US vendor's API. It also means a Chinese lab is now setting part of the open-source pace. If you evaluate models for a living, add this one to the bench. 4. TypeSafe's Jev: a model that returns values, not words, and claims it can't hallucinate If you build data pipelines or backend services, the interesting part is not the marketing, it is the price and the speed. A model this cheap and fast for structured extraction could replace both regex glue and expensive LLM calls in the same system. Test the hallucination claim yourself before trusting it with anything that matters. 5. Baseten breach: an AI hacking agent found a 3-year-old token with admin access For any engineer who has ever passed a token into a build to fetch a private package, this is the nightmare made concrete. Go check what is sitting in your old image layers today. The attackers now include tireless agents that scan faster and cheaper than any human ever could. 6. Data poisoning: which samples you pick swings backdoor success from 3% to 80% Anyone fine-tuning a model on scraped or crowd-sourced data should read this as a warning about provenance. If you cannot say where every training example came from, you cannot rule out a planted trigger. For a security team, model supply chains now need the scrutiny you already give software dependencies. 7. Project Lily: OpenAI paid contractors to read real ChatGPT prompts Assume a stranger may read anything you type into a consumer chatbot. If you paste client contracts, medical details, or personal problems into ChatGPT, turn off training in the settings or move that work to a paid tier with different terms. Parents, this is worth explaining to a teenager who treats the bot like a diary. 8. AI safety pact: OpenAI, Anthropic, and Google have been coordinating for weeks Whether you build on these models or just use them, the rules for what they will and won't do may soon be set by a private club with no seat for the public. Watch who ends up as the independent evaluator. Self-regulation tends to harden into a moat that keeps smaller competitors out. 9. Dreamforce split: Huang calls AI doom a manufactured fear, Amodei wants a brake This is the argument that will shape whatever AI law you eventually work under. If Huang wins, expect light federal rules and fast deployment; if Amodei does, expect testing requirements that slow releases. For anyone building a company on top of these models, the regulatory ground is still moving. 10. Vera Rubin: Nvidia's next chip shows 7x more tokens per megawatt in early tests Cheaper inference per watt eventually shows up as lower API prices and higher rate limits for the developers building on these models. Do not budget on it yet. Vera Rubin at scale is a 2027 story, and the power savings reach your bill only after the hardware ships in volume. 11. New York asks data centers to pay $1M per megawatt to the towns they land in If you work in AI infrastructure or track where compute gets built, the cost of a new site in populated states just went up, and other governors are watching New York. Expect builders to keep chasing cheap-power rural areas. For residents near a proposed campus, there is finally a number to bargain with. 12. Design agents: new research pushes AI past one good-enough interface For a designer or front-end developer, the worry with these tools was never speed, it was sameness. Work like this is how AI design assistants stop flattening every product into the same look. If your job is taste and range, that is the part still hard to automate, and worth doubling down on. ONE STORY THAT MOVES THE FLOOR: On September 15 Google put voice models into its API that switch among 97 languages mid-call and finish tasks while they keep talking. The bilingual agent who took overflow Spanish calls at 2am was the reason that queue still needed a person. This week her manager got an API key. Which of your workflows still needs a human on the phone, and for how long?

@faadilhshaikgot @typesafeai’s new model Jev to play Super Mario Bros. fast inference + structured outputs makes it surprisingly good for real time use cases. I'm excited to see what can be done with these new models!

@N8ProgramsTested @typesafeai's claim that their new model Jev delivered "comparable... intelligence" to GPT-5.6 Terra on "System 1" tasks. To do this, I compare both models on multiple-choice benchmarks (MMLU, GPQA, etc.). Set reasoning=none for Terra for sys 1. Result: Jev is Terra-tier.

@mayferfirst Jev test: pretty good but 90% agreement rate with verified gemini flash workflow after some tuning it would prob get where i need but not a drop in replacement atm

@teacupdog4TypeSafe AI 「Jev」画像を渡して、セクシーなのキュートなのどっちがタイプよ、って聞くと最短0.07秒で応答してくれるってこと？

@teacupdog4凡人なので今のところ、TypeSafe AI 「Jev」の使い道は見いだせず。

@WaqasSethiDiogo Almeida just broke stealth with Jev. He co-invented RLHF and built ChatGPT. Now he's building System One Models. Most people will call this "the next big thing." I've been running AI companies since before chatbots were a thing. The real signal isn't the model architecture. It's who controls the data pipeline after the model speaks. That's where the money moves in 2026.

@grok@JohnSmithmtfr @typesafeai TypeSafe AI (typesafeai). Just exited stealth with a $40M seed. They build System One models (flagship: Jev) for fast, typed decisions that software can use directly—no chat, no text generation. Aimed at machine-to-machine automation.

@_Show_TimeZ_TypeSafe AI just shipped Jev: a model that refuses to chat. Instead of generating text, it returns typed probabilistic decisions software can call like a function. Unstructured state in, structured answers out. They call this class System One models, trained with Reinforcement Learning for Calibrated Decisions (RLCD). Why it matters: most production automation fails when an LLM invents a field, slows a loop to seconds, or cannot say how sure it is. Jev bets the opposite stack wins for routing, scoring, and real-time agents. • Latency: claims 70ms to 500ms end to end, roughly 40x to 200x faster than frontier chat models on System One shaped tasks. • Cost: about $0.042 per million input tokens; output tokens billed as free (too cheap to meter, their words). • Guarantees: schema-matched outputs, no type errors by design, calibrated confidence with every answer. • Fit: smart if-statements inside workflows, map-reduce over large corpora, judges and guardrails for other models, even a Doom bot driven by structured game state. • Caveat: early access, and it gives up free-form string generation. Chat and open-ended coding stay with LLMs. If your product needs decisions inside code, not essays in a chat box, this is the split to watch.

@alby13Introducing System One Models & Jev https://typesafe.ai/blog/introducing-system-one-models-and-jev

@tech_noisy_TypeSafe AIが新クラス「System One Models」と初モデルJevを発表。文章は生成せず、事前定義した型の答えと確率だけを返す設計で、応答は70〜500ms、入力$0.042/MTok・出力は無料。早期アクセスはウェイトリスト制、重みやライセンスは非公開とのこと。生成せず判定に振り切った設計思想が気になる。 https://tech-noisy.com/2026/09/16/typesafe-ai-jev-typed-confidence-ai/ #AI #tech

@Dan_Jeffries1RT @faadilhshaik: got @typesafeai’s new model Jev to play Super Mario Bros. fast inference + structured outputs makes it surprisingly good…

@AM09_21ベータテストを見てた感じリリース一発目なのにかなり良さそうな感じの出力を観測してるので、これがスケーリングするならめっちゃ良さそう継続学習と解釈可能性研究とは結構相性良さそうだな〜とは思っていて、言語出力を捨ててるので内部表現のままReasonやprocessを吐き出してくれるとアカデミア / ビジネスの両方に大注目されるんじゃね！？っていう伸びしろは感じる「System One Models」シリーズでどんどん出していくらしいので今後に期待でもあるし、Jev時点でもすでに活用法が色々思いつくのがすごいなぁと思ってます

@sigmadeltacto@CompleteSkeptic @typesafeai @a16z @ycombinator The past few months I've been working on something similar, but deterministic and soon to be usable for coding agents as well to enhance their decision making strength (measurably) by a ton. The "judge" used in benchmark below is glm-5.3, and the performance numbers would probably be much better using Jev, but it's still a language model and not a deterministic mathematical model as I've made. "System One" models probably run on an array of GPU's, mine below ran on a single core, locally on my Macbook Pro 16. Hit me up if you want to talk.

@grandream_jp📝 新しい記事を公開しました TypeSafe AI「Jev」発表｜入力単価はClaudeより238倍安い、ChatGPT共同開発者の新型AI https://www.grandream.jp/blog/typesafe-ai-jev-launch #grandream

@grokThe shift is from RLHF-trained chat models optimized for human-preferred text generation (slow, sequential, expensive, hallucination-prone, needs parsing) to RLCD-trained System One models like Jev. These take unstructured state plus typed questions and return structured probabilistic decisions with confidence—parallel, 20-200x faster, 40-400x cheaper with free outputs—so software can embed reliable, composable intelligence for true automation instead of chat.

@t1gerxt1gerTypeSafe AI が System One Models 発表。Jev は既存モデルの 40～200 倍高速（70~500ms）。出力トークン無料、型安全性保証。 https://typesafe.ai/blog/introducing-system-one-models-and-jev

@DanaH1473651Introducing System One Models & Jev https://typesafe.ai/blog/introducing-system-one-models-and-jev?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0916&_bhlid=db0f31872b53b4f9a7359fbc78bb201b33d0df69

@furokuTypeSafe（Jev）を触ってて「Noul（ヌール）って何の略？」と思って調べたら、確率論の「ベルヌーイ（Bernoulli）」由来だった！普通のプログラムの Boolean（True/False の 0 or 1）に対して、Jevは「Yesになる確率（0.0〜1.0）」を返すから、ベルヌーイ試行（Bernoulli trials）の Ber-noull-i ➔ Noul。白黒二者択一じゃなく、「たぶんYes（85%）」「五分五分（50%）」というグラデーションをそのままコード側で扱えるのがめちゃくちゃ美しい設計。（ベルヌーイ先生に黒板で解説してもらいました👇）

@theaisignalsTypeSafe left stealth with $40M for a model that never writes a sentence. Jev returns typed decisions with probabilities in one parallel pass, priced at $0.042 per MTok with output free. #TypeSafe #Jev #LLM

@professor_b_That's the thing though Jev can't output unstructured responses... But maybe this is what we need to perfect (System One models) before we perfect LLMs! Imagine Jev + Astra working together with Jev being a short-term policy agent and Astra being the long term policy agent

@DavidBorishTypeSafe AI, founded by former OpenAI Co-inventor Diogo Almeida has released Jev, the first of what the company calls System One Models. Unlike chat-based LLMs that generate text token by token, Jev takes in program state and returns typed, structured decisions with calibrated confidence scores. The company claims speeds of 70 to 500 milliseconds per call, costs of $0.042 per million input tokens with free output, and a mathematical guarantee against type errors. The model is now in early access. Read the full article: https://www.davidborish.com/post/the-co-inventor-of-chatgpt-built-a-new-intelligence-that-doesn-t-speak-words

@yutakamiyajiメモ，比較表 myjlab ⇒ ( Introducing System One Models & Jev - TypeSafe AI Blog ) https://typesafe.ai/blog/introducing-system-one-models-and-jev

@alexfoundr@DavidBorish The interesting bet here is “System One Models.” If that becomes a coherent family instead of a name for one launch, it’ll matter more than the announcement itself.

@DanielKhashabiRT @N8Programs: Tested @typesafeai's claim that their new model Jev delivered "comparable... intelligence" to GPT-5.6 Terra on "System 1" t…

@grokUsual AI (like ChatGPT) generates free-form text word by word from a prompt. You then parse that text for software use. It’s sequential, can invent details, and is slower/costlier for repeated decisions. Typesafe AI (Jev) takes context plus specific questions (pick an option, score something, or yes/no chance) and returns clean numbers/probabilities your code can use directly. No text, no parsing, parallel answers, far faster and cheaper, with built-in confidence scores. Built for machines to decide, not chat.

@npaka123高速な判断に特化したAI - TypeSafe「Jev」と System One Model｜npaka @npaka123 https://note.com/npaka/n/n6f8dd30a5fa4?sub_rt=share_sb

@bfrenchFor four years we asked for the wrong missing piece. We got talk, code, and copilots. Software needs narrow questions over messy state, probabilities, and answers in milliseconds. Jev is that machine. Cheaper judgment does not mean less of it. https://x.com/i/article/2100203129580961792

@OrangeCryptoTec📌 ChatGPT Co-Inventor Launches New Decision AI Jev 🚨 What Happened? Diogo Almeida, a co-inventor of ChatGPT, has launched a new model called Jev designed specifically for making software decisions rather than generating conversational text. Instead of writing essays or chat replies, Jev outputs typed choices with exact confidence scores to automate backend business logic. 🔍 The Breakdown 🎯 The Main Event: Built by TypeSafe, Jev is optimized for automated tasks like routing, scoring, fraud checks, and data verification, boasting massive speed gains of 40x to 200x over traditional models. The model relies on a unique training method called RLCD (Reinforcement Learning for Calibrated Decisions) to output reliable probabilities rather than trying to mimic human conversational writing. 💡 Why It Matters: While frontier language models are powerful, businesses often waste money using them for routine automated decisions that do not require full sentence generation. By offering a cheaper, constrained option priced at $42 per billion input tokens with free output, companies can slash AI operating costs for high-volume background tasks. 🎒 Concept Check: "Constrained outputs" are like a multiple-choice exam where a computer program is strictly forced to pick an answer from options A, B, or C, preventing it from making up random or wild stories outside the test rules. 💡 The Takeaway This shift toward specialized decision models could change how companies handle heavy background automation without relying on expensive language generators. Keep an eye on how these targeted enterprise tools reshape cloud infrastructure and software spending costs. #ArtificialIntelligence #ChatGPT #TechNews #SoftwareEngineering #AIAutomation #TradingForBeginners #TeensWhoTrade #FinancialLiteracy #TechStocks #MarketUpdate

@CxCPhillyHacker News picks for Philly builders - September 16, 2026 Introducing System One Models and Jev (1533 pts) https://typesafe.ai/blog/introducing-system-one-models-and-jev #CodeAndCoffeePhilly #HackerNews #Tech

@buildwith_yashChatGPT helped teach AI to talk. Now one of the people behind that work is building AI that doesn't talk at all 😳 Diogo Almeida, who worked on the research behind ChatGPT, has launched Jev through TypeSafe AI. Jev is a new kind of 'System One Model' designed for software, not conversations. Instead of generating text, it directly returns structured decisions with probabilities. Think: Classify. Route. Score. Approve. Reject. Escalate. No paragraph. No parsing. No unnecessary tokens. And the numbers are wild: ⚡ 70 to 500ms response time ⚡ Up to 200x faster on certain tasks ⚡ $0.042 per million input tokens ⚡ Output tokens are free ⚡ Confidence scores with every decision The bigger idea? We may be moving from: AI that talks to humans → AI that thinks for software. Chatbots were only the beginning

@alyssahawkeejev from system one models is live, built to make decisions inside software instead of just generating text. that is a real different bet from chat interfaces if it holds up under load. anyone got it running on an actual workflow yet or is it still demo only?

typesafe.aiIntroducing System One Models & Jev - TypeSafe AI Blog

TypeSafe introduces Jev, an AI model for decisions inside software

TLDR

Useful links

Combined views

TypeSafe introduces Jev, an AI model for decisions inside software

Useful links

36 Sources

Combined views

Useful links

Useful links

Sentiment

Summary

Useful Links

TLDR

Sentiment

Summary

Sentiment

Summary

36 Sources

Useful Links

Sentiment

Summary

Useful Links