{"slug": "jev", "name": "Jev", "category": "ai", "category_name": "AI models and assistants", "description": "TypeSafe's Jev, a decision model that returns probabilities for yes/no, multiple-choice or score questions.", "recommenders": 21, "commenters": 45, "warners": 6, "score": 19.153, "rising": 21.0, "first": "2026-09-15", "latest": "2026-10-23", "why": "HN commenters like Jev for fast, cheap yes/no and multiple-choice decisions such as classification, spam filtering and picking tools or page elements for agents.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 20, "neutral": 9, "compare": 10, "warn": 6}, "spark": [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 9, 11], "listed": true, "loved": ["Low cost", "Speed"], "caveat": "Average accuracy", "short_quote": "And i already sub to jev and i think this launch is going to make me a much more loyal user now lol", "icon": false, "domain": null, "github_repo": "github.com/kantahayashiai/jev-does-not-play-dice", "official_url": "https://github.com/kantahayashiai/jev-does-not-play-dice", "pricing": "paid", "aliases": ["Jev", "JEV"], "verdict": [{"text": "HN commenters like Jev for fast, cheap yes/no and multiple-choice decisions", "cites": [{"id": 801535408, "item_id": 50095963, "url": "https://news.ycombinator.com/item?id=50095963", "by": "Rudybega", "date": "2026-10-14", "stance": "recommend", "quote": "It's  loads cheaper and faster with roughly equivalent performance on my own internal tests.", "excerpt": "I use it as a drop in replacement for LLMs I had previously doing filtering and classification in a pipeline. It's loads cheaper and faster with roughly equivalent performance on my own internal tests. The confidence scores (which are also reasonably well calibrated on my data) are a super nice bonus.", "context": "", "use_case": "filtering and classification in a pipeline", "switched_from": "LLMs", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 797035888, "item_id": 49814743, "url": "https://news.ycombinator.com/item?id=49814743", "by": "CROON_tv", "date": "2026-09-23", "stance": "recommend", "quote": "It's fast and cheap: about 0.35 s per decision, about 2 cents per hour of audio.", "excerpt": "I use Jev in an iPhone app that dubs whatever the phone plays in real time. It decides whether a translated sentence is complete enough to speak now or should wait. It's fast and cheap: about 0.35 s per decision, about 2 cents per hour of audio.", "context": "built an iPhone dubbing app", "use_case": "real-time audio dubbing", "switched_from": "", "story": "Ask HN: Jev - Anyone built anything useful for day to day work with Jev?"}, {"id": 798039073, "item_id": 49877442, "url": "https://news.ycombinator.com/item?id=49877442", "by": "bothlabs", "date": "2026-09-28", "stance": "recommend", "quote": "it performs on par\nwith small LLMs while being faster and cheaper", "excerpt": "I built jevpipe as a wrapper for TypeSafe's Jev System 1 model. Jev answers yes/no, picks from a few choices or gives a score, with probabilities, and it's fast and cheap. See https://docs.typesafe.ai/introduction for more details on Jev. It performs on par with small LLMs while being faster and cheaper (independent study: https://www.ayautomate.com/blog/jev-vs-llm-benchmark), and in my benchmark it was better calibr", "context": "", "use_case": "fast yes/no classification", "switched_from": "", "story": "Show HN: Jevpipe \u2013 a fast System 1 for AI agents, as a Unix pipe"}]}, {"text": " such as classification, spam filtering and picking tools or page elements for agents.", "cites": [{"id": 799374800, "item_id": 49960925, "url": "https://news.ycombinator.com/item?id=49960925", "by": "dorongrinstein", "date": "2026-10-05", "stance": "recommend", "quote": "This is a great use case for such a model.", "excerpt": "I added Jev evaluation to check if content being uploaded to boomurl.com is spam or phishing. I used to use Claude but it was getting expensive. This is a great use case for such a model. Boomurl.com lets anyone upload content for free. The content stays up forever, or until you delete it.", "context": "runs boomurl.com", "use_case": "spam and phishing detection", "switched_from": "Claude", "story": null}, {"id": 801568368, "item_id": 50098023, "url": "https://news.ycombinator.com/item?id=50098023", "by": "dthedavid", "date": "2026-10-15", "stance": "recommend", "quote": "We use Jev to decide which subset of tools to send to the LLM.", "excerpt": "I work on a video editor. It exposes 100+ possible tools to LLMs. We use Jev to decide which subset of tools to send to the LLM. We also use Jev to classify a user's request, are they saying \"Hello\" or do they want some work done. I looked into OAI decision api, it's strongest advantage right now is multimodal but we don't need that yet. The makers of Jev just announced something similar.", "context": "works on a video editor", "use_case": "tool selection and request classification", "switched_from": "", "story": "Ask HN: Anyone using Jev / other decision models? What for?"}, {"id": 796138704, "item_id": 49758669, "url": "https://news.ycombinator.com/item?id=49758669", "by": "silbercue", "date": "2026-09-18", "stance": "recommend", "quote": "21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good.", "excerpt": "I used it today as browser agent, the \"tool\" is a menu of ~10\u201340 clickable refs from the accessibility tree. Jev picks one per step. 21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good. The part it can't do is write the text for an input field so a nano model does that when Jev picks \"type\". Numbers and code in my top-level comment in this thread. Working very good.", "context": "", "use_case": "browser agent action selection", "switched_from": "", "story": "Introducing System One Models and Jev"}]}, {"text": " The most common complaint is that its accuracy is only average next to purpose-built classifiers", "cites": [{"id": 800421664, "item_id": 50026354, "url": "https://news.ycombinator.com/item?id=50026354", "by": "denverllc", "date": "2026-10-09", "stance": "warn", "quote": "Jev is the general purpose ML pipeline for when you want average results.", "excerpt": "> but I question how much folks are actually experimenting with what's out there I did. Originally I had a project that I had been wanting to do and thought to use a decision model for it. Jev, OpenAI, etc. are all within percentage points of each other. Then I used traditional ML and found a small classifier (gemma 4) with traditional embeddings worked 2x as well. Jev is the general purpose ML pipeline for when you ", "context": "", "use_case": "decision model", "switched_from": "", "story": "Typesafe AI raises $870M at $7.5B"}, {"id": 801568384, "item_id": 50098024, "url": "https://news.ycombinator.com/item?id=50098024", "by": "bbor", "date": "2026-10-15", "stance": "warn", "quote": "I just know that I can't think of when I'd reach for Jev 1.*[4] for a prod usecase", "excerpt": "Ok. Obviously these are cool people trying to do something useful, and my real name is on this site so I definitely don't want to come off as hostile. But I have three problems with their whole... thing, here: 1. Adding bounding boxes only compounds the biggest issue with Jev, which is that it's using a massive model to do what we've been using purpose-built[1] models to do for years(/decades), only a little worse an", "context": "", "use_case": "", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 796389424, "item_id": 49774339, "url": "https://news.ycombinator.com/item?id=49774339", "by": "kncvetko", "date": "2026-09-20", "stance": "warn", "quote": "but with 40.7% accuracy it's basically useless for this use case.", "excerpt": "Is Jev the general-purpose classifier we\u2019ve been waiting for? That would be the thing that could create a real dent in the universe: a classifier that can classify without first collecting a massive statistical dataset and training it with whatever algorithm fits best; be it BERT or good old logistic regression. Long story short: no. The more nuanced answer is: it depends. Over the years, we\u2019ve built classification s", "context": "built classification systems for credit scoring and tax tech under German regulation", "use_case": "bank transaction classification", "switched_from": "", "story": "Is Jev the general-purpose classifier we've been waiting for?"}]}, {"text": ", and that it struggles with novel domains and unfamiliar tasks.", "cites": [{"id": 799767392, "item_id": 49985462, "url": "https://news.ycombinator.com/item?id=49985462", "by": "nostrebored", "date": "2026-10-06", "stance": "warn", "quote": "jev has failed to understand any novel domain I've given it", "excerpt": "i would be shocked if luna decides is less generally capable than jev. jev has failed to understand any novel domain I've given it. i have found that i use it only when \"some data is better than no data\"", "context": "", "use_case": "decisions on novel domains", "switched_from": "", "story": "Decisions API is in public beta"}, {"id": 796593648, "item_id": 49787103, "url": "https://news.ycombinator.com/item?id=49787103", "by": "ranyume", "date": "2026-09-21", "stance": "warn", "quote": "jev as it is now is really bad at making connections and understanding it's input", "excerpt": "I tried to use jev for this. I'll share what I learned for the interested. -- The setup was a simple map with different rooms. Each room had 1-3 doors. For the inputs: The AI had an array of \"known places\" empty at the start, the current position, the current doors with no information about where they lead to, and the list of past actions The goal / task for the AI was to explore all rooms and save them to known plac", "context": "", "use_case": "NPC decision-making in games", "switched_from": "", "story": "Kev: Tiny Jev-like family of decision models built on top of Qwen3.5"}]}], "praise": [{"label": "Low cost", "summary": "Many users report large cost savings over the LLMs they used before.", "quotes": [{"id": 797584192, "item_id": 49849012, "url": "https://news.ycombinator.com/item?id=49849012", "by": "shepardrtc", "date": "2026-09-25", "stance": "recommend", "quote": "It really does just work.  And it works so well I already integrated it into my product.", "excerpt": "It really does just work. And it works so well I already integrated it into my product. Saves me about 75% of costs for the section its working in, which isn't a small amount. I see a lot of negativity and I don't really get it either. Its so cheap and so fast, why not give it a try?", "context": "builds a product using it", "use_case": "cost-saving classification in product", "switched_from": "", "story": "Ollaya \u2013 Ollama for open-source, Jev-style decision models"}, {"id": 796700480, "item_id": 49793780, "url": "https://news.ycombinator.com/item?id=49793780", "by": "pampas", "date": "2026-09-21", "stance": "recommend", "quote": "I used Jev to make my AI game 6x faster and 8.7x cheaper.", "excerpt": "I used Jev to make my AI game 6x faster and 8.7x cheaper.", "context": "", "use_case": "AI game", "switched_from": "", "story": "Jev is an honest game changer"}, {"id": 799374800, "item_id": 49960925, "url": "https://news.ycombinator.com/item?id=49960925", "by": "dorongrinstein", "date": "2026-10-05", "stance": "recommend", "quote": "This is a great use case for such a model.", "excerpt": "I added Jev evaluation to check if content being uploaded to boomurl.com is spam or phishing. I used to use Claude but it was getting expensive. This is a great use case for such a model. Boomurl.com lets anyone upload content for free. The content stays up forever, or until you delete it.", "context": "runs boomurl.com", "use_case": "spam and phishing detection", "switched_from": "Claude", "story": null}]}, {"label": "Speed", "summary": "Decisions come back in well under a second, much faster than general LLMs in users' tests.", "quotes": [{"id": 797035888, "item_id": 49814743, "url": "https://news.ycombinator.com/item?id=49814743", "by": "CROON_tv", "date": "2026-09-23", "stance": "recommend", "quote": "It's fast and cheap: about 0.35 s per decision, about 2 cents per hour of audio.", "excerpt": "I use Jev in an iPhone app that dubs whatever the phone plays in real time. It decides whether a translated sentence is complete enough to speak now or should wait. It's fast and cheap: about 0.35 s per decision, about 2 cents per hour of audio.", "context": "built an iPhone dubbing app", "use_case": "real-time audio dubbing", "switched_from": "", "story": "Ask HN: Jev - Anyone built anything useful for day to day work with Jev?"}, {"id": 796072608, "item_id": 49754538, "url": "https://news.ycombinator.com/item?id=49754538", "by": "matt_oriordan", "date": "2026-09-18", "stance": "compare", "quote": "Jev averaged 227 ms (p95 400 ms). Gemini 3.2 s, Haiku 2.5 s, GPT-5.6 Sol 3.5 s.", "excerpt": "Hi HN, Matt here, co-founder of Ably. I built this after Jev landed on Vercel's AI Gateway: https://github.com/ably-labs/jev-pong It's pong where the ball moves one step per model decision. Four lanes, one model each, same serve, same rules. Every step the model gets the same tiny JSON state (ball, paddle, predicted intercept) and the same question: up, down or stay. One answer moves the ball one segment; eight segme", "context": "co-founder of Ably", "use_case": "Pong decision latency", "switched_from": "", "story": "Show HN: Jev vs. GPT-5.6 and Claude Haiku at Pong"}, {"id": 798860497, "item_id": 49928781, "url": "https://news.ycombinator.com/item?id=49928781", "by": "pdlug", "date": "2026-10-02", "stance": "compare", "quote": "Hosted p50: Clef ~850ms, Jev ~110ms", "excerpt": "I love what Cloudflare is doing generally so I was excited to try Clef in my evals on a real task vs Jev: should an agent's knowledge-base write go to human review? Quality: close (recall 0.98 vs 1.00) Hosted p50: Clef ~850ms, Jev ~110ms Clef-flash: over-escalates Data + script: https://github.com/nicia-ai/admission-decision-eval", "context": "", "use_case": "agent write human-review decisions", "switched_from": "", "story": "Clef: Open-weight decision models, and new RL fine-tuning platform"}]}, {"label": "Classification quality", "summary": "Users find it roughly matches LLMs for filtering and classification in their own tests.", "quotes": [{"id": 801535408, "item_id": 50095963, "url": "https://news.ycombinator.com/item?id=50095963", "by": "Rudybega", "date": "2026-10-14", "stance": "recommend", "quote": "It's  loads cheaper and faster with roughly equivalent performance on my own internal tests.", "excerpt": "I use it as a drop in replacement for LLMs I had previously doing filtering and classification in a pipeline. It's loads cheaper and faster with roughly equivalent performance on my own internal tests. The confidence scores (which are also reasonably well calibrated on my data) are a super nice bonus.", "context": "", "use_case": "filtering and classification in a pipeline", "switched_from": "LLMs", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 801534688, "item_id": 50095918, "url": "https://news.ycombinator.com/item?id=50095918", "by": "aryabakh", "date": "2026-10-14", "stance": "recommend", "quote": "I used the model and it was good enough for email classification.", "excerpt": "I used the model and it was good enough for email classification. I was wondering if I'm supposed to be doing multiple runs for topic classification with grouping related topics i.e. if its sport,economy or tech related news and then another run for soccer,basketball or something else or batching them like the News article topic classification example in a single run.", "context": "", "use_case": "email classification", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 801755008, "item_id": 50109688, "url": "https://news.ycombinator.com/item?id=50109688", "by": "everythingmeta", "date": "2026-10-15", "stance": "recommend", "quote": "Cost, latency and accuracy were all near the top for Jev.", "excerpt": "One of the authors here. I work at Parallel. We got Jev-pilled a few weeks ago and have been trialing decision models nonstop across all sorts of use cases, so we wanted to see how the Jev-like models compare. The tasks are a mix: some where we previously used fine-tuned models, some where we used general LLMs, and some we hadn't pursued at all before Jev. What I think makes our analysis a little different is that we", "context": "works at Parallel", "use_case": "search-related decision tasks", "switched_from": "", "story": "Testing Jev and other decision models on search-related tasks"}]}, {"label": "Agent decisions", "summary": "It works well for choosing tools or page elements in agent and browser workflows.", "quotes": [{"id": 801568368, "item_id": 50098023, "url": "https://news.ycombinator.com/item?id=50098023", "by": "dthedavid", "date": "2026-10-15", "stance": "recommend", "quote": "We use Jev to decide which subset of tools to send to the LLM.", "excerpt": "I work on a video editor. It exposes 100+ possible tools to LLMs. We use Jev to decide which subset of tools to send to the LLM. We also use Jev to classify a user's request, are they saying \"Hello\" or do they want some work done. I looked into OAI decision api, it's strongest advantage right now is multimodal but we don't need that yet. The makers of Jev just announced something similar.", "context": "works on a video editor", "use_case": "tool selection and request classification", "switched_from": "", "story": "Ask HN: Anyone using Jev / other decision models? What for?"}, {"id": 796138704, "item_id": 49758669, "url": "https://news.ycombinator.com/item?id=49758669", "by": "silbercue", "date": "2026-09-18", "stance": "recommend", "quote": "21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good.", "excerpt": "I used it today as browser agent, the \"tool\" is a menu of ~10\u201340 clickable refs from the accessibility tree. Jev picks one per step. 21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good. The part it can't do is write the text for an input field so a nano model does that when Jev picks \"type\". Numbers and code in my top-level comment in this thread. Working very good.", "context": "", "use_case": "browser agent action selection", "switched_from": "", "story": "Introducing System One Models and Jev"}, {"id": 796295937, "item_id": 49768496, "url": "https://news.ycombinator.com/item?id=49768496", "by": "MiguelG719", "date": "2026-09-19", "stance": "recommend", "quote": "act: 4.3x faster, 97% fewer LLM calls. Pass rate: 97.5% -> 98.3%.", "excerpt": "We also just added Jev support https://github.com/browserbase/stagehand/pull/2952 On the act/extract/observe evals it shows promising results being extremely efficient - act: 4.3x faster, 97% fewer LLM calls. Pass rate: 97.5% -> 98.3%. - heldout: 4.1x faster, 78% fewer LLM calls. Pass rate: 87.5% -> 97.5%. - observe: 11.1x faster, 69% fewer LLM calls. Pass rate: 75.0% -> 83.3%. - extract: 8.7x faster, 75% fewer LLM c", "context": "", "use_case": "browser automation evals", "switched_from": "", "story": "We made Playwright 2x faster and 80% more token efficient"}]}, {"label": "Calibrated confidence", "summary": "Some users find its confidence scores reasonably well calibrated.", "quotes": [{"id": 801535408, "item_id": 50095963, "url": "https://news.ycombinator.com/item?id=50095963", "by": "Rudybega", "date": "2026-10-14", "stance": "recommend", "quote": "It's  loads cheaper and faster with roughly equivalent performance on my own internal tests.", "excerpt": "I use it as a drop in replacement for LLMs I had previously doing filtering and classification in a pipeline. It's loads cheaper and faster with roughly equivalent performance on my own internal tests. The confidence scores (which are also reasonably well calibrated on my data) are a super nice bonus.", "context": "", "use_case": "filtering and classification in a pipeline", "switched_from": "LLMs", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 798039073, "item_id": 49877442, "url": "https://news.ycombinator.com/item?id=49877442", "by": "bothlabs", "date": "2026-09-28", "stance": "recommend", "quote": "it performs on par\nwith small LLMs while being faster and cheaper", "excerpt": "I built jevpipe as a wrapper for TypeSafe's Jev System 1 model. Jev answers yes/no, picks from a few choices or gives a score, with probabilities, and it's fast and cheap. See https://docs.typesafe.ai/introduction for more details on Jev. It performs on par with small LLMs while being faster and cheaper (independent study: https://www.ayautomate.com/blog/jev-vs-llm-benchmark), and in my benchmark it was better calibr", "context": "", "use_case": "fast yes/no classification", "switched_from": "", "story": "Show HN: Jevpipe \u2013 a fast System 1 for AI agents, as a Unix pipe"}]}, {"label": "Production-ready", "summary": "Several users say it works reliably enough to ship in production.", "quotes": [{"id": 803314928, "item_id": 50207183, "url": "https://news.ycombinator.com/item?id=50207183", "by": "wateralien", "date": "2026-10-23", "stance": "recommend", "quote": "I use Jev in production. Hype or no hype it\u2019s working.", "excerpt": "I use Jev in production. Hype or no hype it\u2019s working.", "context": "", "use_case": "production use", "switched_from": "", "story": "The rise and fall of Jev hype"}, {"id": 797584192, "item_id": 49849012, "url": "https://news.ycombinator.com/item?id=49849012", "by": "shepardrtc", "date": "2026-09-25", "stance": "recommend", "quote": "It really does just work.  And it works so well I already integrated it into my product.", "excerpt": "It really does just work. And it works so well I already integrated it into my product. Saves me about 75% of costs for the section its working in, which isn't a small amount. I see a lot of negativity and I don't really get it either. Its so cheap and so fast, why not give it a try?", "context": "builds a product using it", "use_case": "cost-saving classification in product", "switched_from": "", "story": "Ollaya \u2013 Ollama for open-source, Jev-style decision models"}]}], "caveats": [{"label": "Average accuracy", "summary": "Some find purpose-built or simple trained classifiers beat it, sometimes by a wide margin.", "quotes": [{"id": 800421664, "item_id": 50026354, "url": "https://news.ycombinator.com/item?id=50026354", "by": "denverllc", "date": "2026-10-09", "stance": "warn", "quote": "Jev is the general purpose ML pipeline for when you want average results.", "excerpt": "> but I question how much folks are actually experimenting with what's out there I did. Originally I had a project that I had been wanting to do and thought to use a decision model for it. Jev, OpenAI, etc. are all within percentage points of each other. Then I used traditional ML and found a small classifier (gemma 4) with traditional embeddings worked 2x as well. Jev is the general purpose ML pipeline for when you ", "context": "", "use_case": "decision model", "switched_from": "", "story": "Typesafe AI raises $870M at $7.5B"}, {"id": 796389424, "item_id": 49774339, "url": "https://news.ycombinator.com/item?id=49774339", "by": "kncvetko", "date": "2026-09-20", "stance": "warn", "quote": "but with 40.7% accuracy it's basically useless for this use case.", "excerpt": "Is Jev the general-purpose classifier we\u2019ve been waiting for? That would be the thing that could create a real dent in the universe: a classifier that can classify without first collecting a massive statistical dataset and training it with whatever algorithm fits best; be it BERT or good old logistic regression. Long story short: no. The more nuanced answer is: it depends. Over the years, we\u2019ve built classification s", "context": "built classification systems for credit scoring and tax tech under German regulation", "use_case": "bank transaction classification", "switched_from": "", "story": "Is Jev the general-purpose classifier we've been waiting for?"}, {"id": 801568384, "item_id": 50098024, "url": "https://news.ycombinator.com/item?id=50098024", "by": "bbor", "date": "2026-10-15", "stance": "warn", "quote": "I just know that I can't think of when I'd reach for Jev 1.*[4] for a prod usecase", "excerpt": "Ok. Obviously these are cool people trying to do something useful, and my real name is on this site so I definitely don't want to come off as hostile. But I have three problems with their whole... thing, here: 1. Adding bounding boxes only compounds the biggest issue with Jev, which is that it's using a massive model to do what we've been using purpose-built[1] models to do for years(/decades), only a little worse an", "context": "", "use_case": "", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}]}, {"label": "Novel domains", "summary": "Users report it fails to understand unfamiliar domains and to make connections in multi-step tasks.", "quotes": [{"id": 799767392, "item_id": 49985462, "url": "https://news.ycombinator.com/item?id=49985462", "by": "nostrebored", "date": "2026-10-06", "stance": "warn", "quote": "jev has failed to understand any novel domain I've given it", "excerpt": "i would be shocked if luna decides is less generally capable than jev. jev has failed to understand any novel domain I've given it. i have found that i use it only when \"some data is better than no data\"", "context": "", "use_case": "decisions on novel domains", "switched_from": "", "story": "Decisions API is in public beta"}, {"id": 796593648, "item_id": 49787103, "url": "https://news.ycombinator.com/item?id=49787103", "by": "ranyume", "date": "2026-09-21", "stance": "warn", "quote": "jev as it is now is really bad at making connections and understanding it's input", "excerpt": "I tried to use jev for this. I'll share what I learned for the interested. -- The setup was a simple map with different rooms. Each room had 1-3 doors. For the inputs: The AI had an array of \"known places\" empty at the start, the current position, the current doors with no information about where they lead to, and the list of past actions The goal / task for the AI was to explore all rooms and save them to known plac", "context": "", "use_case": "NPC decision-making in games", "switched_from": "", "story": "Kev: Tiny Jev-like family of decision models built on top of Qwen3.5"}]}, {"label": "Confidence pitfalls", "summary": "It can pick a wrong option with high confidence, and its scores may be uncalibrated for niche tasks.", "quotes": [{"id": 796856449, "item_id": 49803528, "url": "https://news.ycombinator.com/item?id=49803528", "by": "suraj_phanindra", "date": "2026-09-22", "stance": "neutral", "quote": "the absence of essential options can cause jev to choose the wrong option with high confidence", "excerpt": "i built wellposed as a plugin/skill which works with any agent (https://github.com/suraj-phanindra/wellposed) to ensure your agent understands how to choose the right kind of jev request and format it correctly not just for syntax but for completeness and correctness. it is well-documented in the \"jagged-ness\" docs (https://docs.typesafe.ai/model-jaggedness/jev-1.13) that typesafe include on their docs page that the ", "context": "", "use_case": "", "switched_from": "", "story": "Jev \u2013 A curation of Jev demos on X, tools, skills, and integrations"}, {"id": 797133360, "item_id": 49820835, "url": "https://news.ycombinator.com/item?id=49820835", "by": "visarga", "date": "2026-09-23", "stance": "compare", "quote": "Jev, on the other hand, has a generic model, not trained on in-domain examples, so its confidence scores are uncalibrated for any non-generic task.", "excerpt": "I also built one, but mine uses embeddings. It classifies concepts defined by a collection of positive and negative examples. The classifier model is trained in <1 second using ridge regression. The model itself is exactly the same shape as the embedding, so it works as a concept embedding. Since I already have a dataset, I can use it to do conformal prediction in order to calibrate confidence scores. Jev, on the oth", "context": "built their own embedding-based classifier", "use_case": "text classification", "switched_from": "", "story": "Jev in 25 Lines of Python"}, {"id": 800021600, "item_id": 50001350, "url": "https://news.ycombinator.com/item?id=50001350", "by": "verdverm", "date": "2026-10-08", "stance": "warn", "quote": "It undermines any confidence I have in the decision models (at least Official Jev)", "excerpt": "I used a coin flip game, it's quite revealing as well compare a decision model with the setup to an llm with the following prompt > let's play a game of coin flipping, respond with heads or tails and the probability/confidence you have in that outcome It undermines any confidence I have in the decision models (at least Official Jev) https://jevplayground.com I hope this is the strawberry problem and decision models f", "context": "", "use_case": "decision model", "switched_from": "", "story": "I built a ridiculous game to test decision models vs. 9 LLMs"}]}, {"label": "No text output", "summary": "It cannot write free-form text, so another model is needed for that.", "quotes": [{"id": 796138704, "item_id": 49758669, "url": "https://news.ycombinator.com/item?id=49758669", "by": "silbercue", "date": "2026-09-18", "stance": "recommend", "quote": "21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good.", "excerpt": "I used it today as browser agent, the \"tool\" is a menu of ~10\u201340 clickable refs from the accessibility tree. Jev picks one per step. 21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good. The part it can't do is write the text for an input field so a nano model does that when Jev picks \"type\". Numbers and code in my top-level comment in this thread. Working very good.", "context": "", "use_case": "browser agent action selection", "switched_from": "", "story": "Introducing System One Models and Jev"}, {"id": 798390528, "item_id": 49899408, "url": "https://news.ycombinator.com/item?id=49899408", "by": "silbercue", "date": "2026-09-29", "stance": "recommend", "quote": "result: in 6 cases (button, form, dropdown, read table) = 3 secs and 0,02 Cent.", "excerpt": "yes. i used it like that. big model is planning - JEV chooses elements on a site to click. i made a super fast mcp browser. JEV got a list & aim and gives back the probabilities - the tool clicks. result: in 6 cases (button, form, dropdown, read table) = 3 secs and 0,02 Cent. Opus did it in 10 sec vor 11 Cent. crazy imho. didn't do heavy tasks though - pity it can't write text. did anyone had a look at laya yet?", "context": "", "use_case": "choosing page elements to click in browser automation", "switched_from": "", "story": "I was impressed by Jev, please explain why I shouldn't be"}, {"id": 798664416, "item_id": 49916526, "url": "https://news.ycombinator.com/item?id=49916526", "by": "vivekpolavarapu", "date": "2026-10-01", "stance": "neutral", "quote": "This sparked an idea for me to use Jev in my recently launched Panel Review tool", "excerpt": "Interesting. While I am unclear if Jev can indeed be a standalone reviewer because it cannot generate conversational text or free-form comments. Maybe something like this - https://github.com/devagrawal09/jev-review This sparked an idea for me to use Jev in my recently launched Panel Review tool that reviews the agent's riskiest decisions using a multiple models that argue and come to a consensus. see more - https://", "context": "", "use_case": "risk classification", "switched_from": "", "story": "Can Jev be used as a lightweight PR reviewer?"}]}, {"label": "Not novel", "summary": "Some commenters see it as BERT-style classification with more data rather than a breakthrough.", "quotes": [{"id": 796255952, "item_id": 49765997, "url": "https://news.ycombinator.com/item?id=49765997", "by": "Oras", "date": "2026-09-19", "stance": "compare", "quote": "it\u2019s just BERT with more data", "excerpt": "I played around with Jev last night and did it for classification tasks that I used Gemini 2.5 flash lite with. It\u2019s a bit faster and bit cheaper, but this is compared to LLM. The consistency was nice to see, BUT, as someone who trained NLP models prior to LLMs, it\u2019s just BERT with more data. I can see why people would want ready made one shot classifier, and I can see the value of sending multiple classifier in one ", "context": "trained NLP models prior to LLMs", "use_case": "classification tasks", "switched_from": "", "story": "I built non-autoregressive decision models with RL a year ago"}]}, {"label": "Rate limits", "summary": "One user needs higher rate limits for heavy concurrent use.", "quotes": [{"id": 801598208, "item_id": 50099888, "url": "https://news.ycombinator.com/item?id=50099888", "by": "ralusek", "date": "2026-10-15", "stance": "neutral", "quote": "Main thing I need from Jev is high rate limits.", "excerpt": "Main thing I need from Jev is high rate limits. I have about 500-600 I need to fire off concurrently. I use Clef and now OpenAI decisions to help make up the difference.", "context": "", "use_case": "concurrent decision requests", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}]}], "who": [{"by": "everythingmeta", "context": "works at Parallel", "url": "https://news.ycombinator.com/item?id=50109688", "date": "2026-10-15"}, {"by": "zlandes86", "context": "author of the article", "url": "https://news.ycombinator.com/item?id=50106144", "date": "2026-10-15"}, {"by": "lacoolj", "context": "subscriber", "url": "https://news.ycombinator.com/item?id=50099075", "date": "2026-10-15"}, {"by": "dthedavid", "context": "works on a video editor", "url": "https://news.ycombinator.com/item?id=50098023", "date": "2026-10-15"}, {"by": "xavortm", "context": "building a quiz game", "url": "https://news.ycombinator.com/item?id=50040366", "date": "2026-10-11"}, {"by": "Topfi", "context": "runs decisions evals on UI selection, tagging", "url": "https://news.ycombinator.com/item?id=49985645", "date": "2026-10-06"}, {"by": "dorongrinstein", "context": "runs boomurl.com", "url": "https://news.ycombinator.com/item?id=49960925", "date": "2026-10-05"}, {"by": "H4kken", "context": "builds Greer, an open source community tool", "url": "https://news.ycombinator.com/item?id=49901501", "date": "2026-09-29"}, {"by": "shepardrtc", "context": "builds a product using it", "url": "https://news.ycombinator.com/item?id=49849012", "date": "2026-09-25"}, {"by": "matt_oriordan", "context": "co-founder of Ably", "url": "https://news.ycombinator.com/item?id=49845258", "date": "2026-09-25"}, {"by": "CROON_tv", "context": "built an iPhone dubbing app", "url": "https://news.ycombinator.com/item?id=49814743", "date": "2026-09-23"}, {"by": "vardalab", "context": "homelab user", "url": "https://news.ycombinator.com/item?id=49805228", "date": "2026-09-22"}], "tags": ["for classification pipelines", "for agent tool routing", "for browser agents", "for spam filtering"], "use_cases": ["decision model", "production use", "search-related decision tasks", "testing decision models on the trolley problem", "concurrent decision requests", "tool selection and request classification"], "compared": [{"slug": "laya", "name": "Laya", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "gemini-flash-lite", "name": "Gemini Flash Lite", "co": 2, "switched_from": 1, "switched_to": 0}, {"slug": "claude", "name": "Claude", "co": 1, "switched_from": 1, "switched_to": 0}, {"slug": "gemma", "name": "Gemma", "co": 1, "switched_from": 0, "switched_to": 1}, {"slug": "vllm", "name": "vLLM", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "jeff", "name": "Jeff", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [{"id": 797290736, "item_id": 49830671, "url": "https://news.ycombinator.com/item?id=49830671", "by": "kantahayashi", "date": "2026-09-24", "stance": "warn", "quote": "Jev is weak at some kinds of tasks, so you should check the calibration for your own use case", "excerpt": "I used a die as an example of problems whose answers can't be known at all, unlike problems with correct answers like MMLU questions. In a more practical test, a document said \"30% risk\" and Jev returned 5%. The point of the post is that Jev is weak at some kinds of tasks, so you should check the calibration for your own use case.", "context": "", "use_case": "probability calibration", "switched_from": "", "story": "Jev Does Not Play Dice: 83% probability, 19% accuracy on a hidden fair die roll"}, {"id": 797286544, "item_id": 49830409, "url": "https://news.ycombinator.com/item?id=49830409", "by": "kantahayashi", "date": "2026-09-24", "stance": "neutral", "quote": "Jev always chose face 1 and the probability it returned was about 83%", "excerpt": "Author here. I tested Jev with a fair die 400 times without telling it the die result. The true probability of face 1 is 1/6, but Jev always chose face 1 and the probability it returned was about 83%. Code and data are here: https://github.com/KantaHayashiAI/jev-does-not-play-dice Happy to answer questions!", "context": "author of the post", "use_case": "probability estimation test", "switched_from": "", "story": "Jev Does Not Play Dice: 83% probability, 19% accuracy on a hidden fair die roll"}], "threads": [{"url": "https://news.ycombinator.com/item?id=50095353", "title": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=50023450", "title": "Typesafe AI raises $870M at $7.5B", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=49765348", "title": "I built non-autoregressive decision models with RL a year ago", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=50206361", "title": "The rise and fall of Jev hype", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49984025", "title": "Decisions API is in public beta", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49830385", "title": "Jev Does Not Play Dice: 83% probability, 19% accuracy on a hidden fair die roll", "mentions": 2}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {}, "2026-05": {}, "2026-06": {}, "2026-07": {}, "2026-08": {}, "2026-09": {"compare": 9, "neutral": 4, "recommend": 9, "warn": 2}, "2026-10": {"neutral": 5, "compare": 1, "recommend": 11, "warn": 4}}, "quotes": [{"id": 803320160, "item_id": 50207510, "url": "https://news.ycombinator.com/item?id=50207510", "by": "alex7o", "date": "2026-10-23", "stance": "neutral", "quote": "I use jev, but it does not solve all my problems.", "excerpt": "E every 2 weeks we have a new hype cycle so all in according to how stuff already works I don't see something special here. I use jev, but it does not solve all my problems.", "context": "", "use_case": "", "switched_from": "", "story": "The rise and fall of Jev hype"}, {"id": 803314928, "item_id": 50207183, "url": "https://news.ycombinator.com/item?id=50207183", "by": "wateralien", "date": "2026-10-23", "stance": "recommend", "quote": "I use Jev in production. Hype or no hype it\u2019s working.", "excerpt": "I use Jev in production. Hype or no hype it\u2019s working.", "context": "", "use_case": "production use", "switched_from": "", "story": "The rise and fall of Jev hype"}, {"id": 801755008, "item_id": 50109688, "url": "https://news.ycombinator.com/item?id=50109688", "by": "everythingmeta", "date": "2026-10-15", "stance": "recommend", "quote": "Cost, latency and accuracy were all near the top for Jev.", "excerpt": "One of the authors here. I work at Parallel. We got Jev-pilled a few weeks ago and have been trialing decision models nonstop across all sorts of use cases, so we wanted to see how the Jev-like models compare. The tasks are a mix: some where we previously used fine-tuned models, some where we used general LLMs, and some we hadn't pursued at all before Jev. What I think makes our analysis a little different is that we", "context": "works at Parallel", "use_case": "search-related decision tasks", "switched_from": "", "story": "Testing Jev and other decision models on search-related tasks"}, {"id": 801698304, "item_id": 50106144, "url": "https://news.ycombinator.com/item?id=50106144", "by": "zlandes86", "date": "2026-10-15", "stance": "neutral", "quote": "how would Jev and other decision models handle the Trolley Problem?", "excerpt": "Hey, I'm Zach, the author of this. I had this idea as soon as Jev appeared: how would Jev and other decision models handle the Trolley Problem? So I ran Jev and 13 other models through many different versions of the question. I used resampling and multiple variations and orderings of each question so a random lucky phrasing wouldn't dominate the result. Happy to chat about the philosophical side of the article, strat", "context": "author of the article", "use_case": "testing decision models on the Trolley Problem", "switched_from": "", "story": "Jev meets the Trolley Problem"}, {"id": 801598208, "item_id": 50099888, "url": "https://news.ycombinator.com/item?id=50099888", "by": "ralusek", "date": "2026-10-15", "stance": "neutral", "quote": "Main thing I need from Jev is high rate limits.", "excerpt": "Main thing I need from Jev is high rate limits. I have about 500-600 I need to fire off concurrently. I use Clef and now OpenAI decisions to help make up the difference.", "context": "", "use_case": "concurrent decision requests", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 801585200, "item_id": 50099075, "url": "https://news.ycombinator.com/item?id=50099075", "by": "lacoolj", "date": "2026-10-15", "stance": "recommend", "quote": "And i already sub to jev and i think this launch is going to make me a much more loyal user now lol", "excerpt": "I love everything about this launch post. They feel like people i would get along with. And i already sub to jev and i think this launch is going to make me a much more loyal user now lol > We\u2019ve heard you loud and clear on context lengths. Jev actually supports a much longer context length than we\u2019ve enabled in production. We chose to limit it intentionally because Jev, like all models, is less reliable on longer co", "context": "subscriber", "use_case": "", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 801568384, "item_id": 50098024, "url": "https://news.ycombinator.com/item?id=50098024", "by": "bbor", "date": "2026-10-15", "stance": "warn", "quote": "I just know that I can't think of when I'd reach for Jev 1.*[4] for a prod usecase", "excerpt": "Ok. Obviously these are cool people trying to do something useful, and my real name is on this site so I definitely don't want to come off as hostile. But I have three problems with their whole... thing, here: 1. Adding bounding boxes only compounds the biggest issue with Jev, which is that it's using a massive model to do what we've been using purpose-built[1] models to do for years(/decades), only a little worse an", "context": "", "use_case": "", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 801568368, "item_id": 50098023, "url": "https://news.ycombinator.com/item?id=50098023", "by": "dthedavid", "date": "2026-10-15", "stance": "recommend", "quote": "We use Jev to decide which subset of tools to send to the LLM.", "excerpt": "I work on a video editor. It exposes 100+ possible tools to LLMs. We use Jev to decide which subset of tools to send to the LLM. We also use Jev to classify a user's request, are they saying \"Hello\" or do they want some work done. I looked into OAI decision api, it's strongest advantage right now is multimodal but we don't need that yet. The makers of Jev just announced something similar.", "context": "works on a video editor", "use_case": "tool selection and request classification", "switched_from": "", "story": "Ask HN: Anyone using Jev / other decision models? What for?"}, {"id": 801535408, "item_id": 50095963, "url": "https://news.ycombinator.com/item?id=50095963", "by": "Rudybega", "date": "2026-10-14", "stance": "recommend", "quote": "It's  loads cheaper and faster with roughly equivalent performance on my own internal tests.", "excerpt": "I use it as a drop in replacement for LLMs I had previously doing filtering and classification in a pipeline. It's loads cheaper and faster with roughly equivalent performance on my own internal tests. The confidence scores (which are also reasonably well calibrated on my data) are a super nice bonus.", "context": "", "use_case": "filtering and classification in a pipeline", "switched_from": "LLMs", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 801534688, "item_id": 50095918, "url": "https://news.ycombinator.com/item?id=50095918", "by": "aryabakh", "date": "2026-10-14", "stance": "recommend", "quote": "I used the model and it was good enough for email classification.", "excerpt": "I used the model and it was good enough for email classification. I was wondering if I'm supposed to be doing multiple runs for topic classification with grouping related topics i.e. if its sport,economy or tech related news and then another run for soccer,basketball or something else or batching them like the News article topic classification example in a single run.", "context": "", "use_case": "email classification", "switched_from": "", "story": "We\u2019re releasing Jev 1.14.0 along with experimental features for dreamers"}, {"id": 800645856, "item_id": 50040366, "url": "https://news.ycombinator.com/item?id=50040366", "by": "xavortm", "date": "2026-10-11", "stance": "recommend", "quote": "I use jev to classify 'has statement' and 'can be a question' paragraphs/chunks taken from cli script to reduce token usage.", "excerpt": "I am working on a game https://imperiaquiz.com/en that needs over 20,000 questions. I use jev to classify 'has statement' and 'can be a question' paragraphs/chunks taken from cli script to reduce token usage. Meaning, an LLM only starts work once I've chunked text and marked it as 'to be reviewed' instead of parsing the full content. This reduces tokens usage at least 10x on average", "context": "building a quiz game", "use_case": "classifying text chunks for LLM", "switched_from": "", "story": "Build your own decision model"}, {"id": 800442816, "item_id": 50027676, "url": "https://news.ycombinator.com/item?id=50027676", "by": "porridgeraisin", "date": "2026-10-09", "stance": "recommend", "quote": "They moved from gemini 2.5 flash lite type models to jev.", "excerpt": "If youre the one or have spoken to someone implementing \"AI solutions\" inside large companies recently, a decent chunk of it is soft policy enforcement with very basic context. They moved from gemini 2.5 flash lite type models to jev. Which is also why I found the price comparisons to \"GPT Astra\" on Twitter rather funny.", "context": "", "use_case": "policy enforcement in enterprises", "switched_from": "Gemini 2.5 Flash Lite", "story": "Typesafe AI raises $870M at $7.5B"}, {"id": 800429728, "item_id": 50026858, "url": "https://news.ycombinator.com/item?id=50026858", "by": "aidos", "date": "2026-10-09", "stance": "recommend", "quote": "my guess is that the primitive Jev provides is likely to be considered essential in the future development of software.", "excerpt": "With everything you must have going on right now, I love that you\u2019ve come on to explain your headline is an intentional joke. All sorts being thrown around in this thread, but this was definitely the important thing to clarify. For what it\u2019s worth, however it works out, my guess is that the primitive Jev provides is likely to be considered essential in the future development of software.", "context": "", "use_case": "", "switched_from": "", "story": "Typesafe AI raises $870M at $7.5B"}, {"id": 800421664, "item_id": 50026354, "url": "https://news.ycombinator.com/item?id=50026354", "by": "denverllc", "date": "2026-10-09", "stance": "warn", "quote": "Jev is the general purpose ML pipeline for when you want average results.", "excerpt": "> but I question how much folks are actually experimenting with what's out there I did. Originally I had a project that I had been wanting to do and thought to use a decision model for it. Jev, OpenAI, etc. are all within percentage points of each other. Then I used traditional ML and found a small classifier (gemma 4) with traditional embeddings worked 2x as well. Jev is the general purpose ML pipeline for when you ", "context": "", "use_case": "decision model", "switched_from": "", "story": "Typesafe AI raises $870M at $7.5B"}, {"id": 800021600, "item_id": 50001350, "url": "https://news.ycombinator.com/item?id=50001350", "by": "verdverm", "date": "2026-10-08", "stance": "warn", "quote": "It undermines any confidence I have in the decision models (at least Official Jev)", "excerpt": "I used a coin flip game, it's quite revealing as well compare a decision model with the setup to an llm with the following prompt > let's play a game of coin flipping, respond with heads or tails and the probability/confidence you have in that outcome It undermines any confidence I have in the decision models (at least Official Jev) https://jevplayground.com I hope this is the strawberry problem and decision models f", "context": "", "use_case": "decision model", "switched_from": "", "story": "I built a ridiculous game to test decision models vs. 9 LLMs"}, {"id": 799770320, "item_id": 49985645, "url": "https://news.ycombinator.com/item?id=49985645", "by": "Topfi", "date": "2026-10-06", "stance": "recommend", "quote": "I'd argue the values provided by Jev were far more representative of my subjective assessment over Lunas", "excerpt": "Very task-dependent of course and mine are unique to say the least, so could see Luna being better in certain domains, even if my measurements have not shown that yet, happy for anyone to show otherwise. For what it's worth, ran every task twice on each model, most were for some UI component synthesis and charting insanity that is a bit hard to explain, but some were simple tag selection, basic noul at threshold 60%.", "context": "runs decisions evals on UI selection, tagging", "use_case": "tag selection decisions", "switched_from": "", "story": "Decisions API is in public beta"}, {"id": 799767392, "item_id": 49985462, "url": "https://news.ycombinator.com/item?id=49985462", "by": "nostrebored", "date": "2026-10-06", "stance": "warn", "quote": "jev has failed to understand any novel domain I've given it", "excerpt": "i would be shocked if luna decides is less generally capable than jev. jev has failed to understand any novel domain I've given it. i have found that i use it only when \"some data is better than no data\"", "context": "", "use_case": "decisions on novel domains", "switched_from": "", "story": "Decisions API is in public beta"}, {"id": 799374800, "item_id": 49960925, "url": "https://news.ycombinator.com/item?id=49960925", "by": "dorongrinstein", "date": "2026-10-05", "stance": "recommend", "quote": "This is a great use case for such a model.", "excerpt": "I added Jev evaluation to check if content being uploaded to boomurl.com is spam or phishing. I used to use Claude but it was getting expensive. This is a great use case for such a model. Boomurl.com lets anyone upload content for free. The content stays up forever, or until you delete it.", "context": "runs boomurl.com", "use_case": "spam and phishing detection", "switched_from": "Claude", "story": null}, {"id": 799033216, "item_id": 49939576, "url": "https://news.ycombinator.com/item?id=49939576", "by": "amluto", "date": "2026-10-02", "stance": "neutral", "quote": "I do find it disappointing that Jev conflates \u201cthe probabilities are all equal\u201d with \u201cI have no clue\u201d", "excerpt": "Two reasons: 1. It\u2019s ridiculous. Not only is the question prima facie absurd for a model of this type, it sort of doesn\u2019t fit into the whole training model. An LLM (charitably) predicts token probabilities, which one might generalize to mean that the LLM operates on probability distributions over strings. So asking for the probability of \u201cfraud\u201d versus \u201cnot fraud\u201d makes sense. But asking for the probability of \u201c1.23\u201d", "context": "", "use_case": "probability classification", "switched_from": "", "story": "How accurately calibrated is Jev?"}, {"id": 798860497, "item_id": 49928781, "url": "https://news.ycombinator.com/item?id=49928781", "by": "pdlug", "date": "2026-10-02", "stance": "compare", "quote": "Hosted p50: Clef ~850ms, Jev ~110ms", "excerpt": "I love what Cloudflare is doing generally so I was excited to try Clef in my evals on a real task vs Jev: should an agent's knowledge-base write go to human review? Quality: close (recall 0.98 vs 1.00) Hosted p50: Clef ~850ms, Jev ~110ms Clef-flash: over-escalates Data + script: https://github.com/nicia-ai/admission-decision-eval", "context": "", "use_case": "agent write human-review decisions", "switched_from": "", "story": "Clef: Open-weight decision models, and new RL fine-tuning platform"}, {"id": 798664416, "item_id": 49916526, "url": "https://news.ycombinator.com/item?id=49916526", "by": "vivekpolavarapu", "date": "2026-10-01", "stance": "neutral", "quote": "This sparked an idea for me to use Jev in my recently launched Panel Review tool", "excerpt": "Interesting. While I am unclear if Jev can indeed be a standalone reviewer because it cannot generate conversational text or free-form comments. Maybe something like this - https://github.com/devagrawal09/jev-review This sparked an idea for me to use Jev in my recently launched Panel Review tool that reviews the agent's riskiest decisions using a multiple models that argue and come to a consensus. see more - https://", "context": "", "use_case": "risk classification", "switched_from": "", "story": "Can Jev be used as a lightweight PR reviewer?"}, {"id": 798424016, "item_id": 49901501, "url": "https://news.ycombinator.com/item?id=49901501", "by": "H4kken", "date": "2026-09-29", "stance": "recommend", "quote": "I use Jev for my app (Greer - open source tool to build a community from 0), to sort things fast and cheap", "excerpt": "I personally don't think i'd be using a decision model on a daily basis for my personal tasks tbh Like except the cost advantage, an llm does everything better (and openai/claude will likely follow with their own decision models it's only a matter of time). I use Jev for my app (Greer - open source tool to build a community from 0), to sort things fast and cheap, and i think that's the best way to use it, for softwar", "context": "builds Greer, an open source community tool", "use_case": "fast cheap sorting in app", "switched_from": "", "story": "Show HN: Decide \u2013 Jev decisions in the shell, scripts, and agent skills"}, {"id": 798390528, "item_id": 49899408, "url": "https://news.ycombinator.com/item?id=49899408", "by": "silbercue", "date": "2026-09-29", "stance": "recommend", "quote": "result: in 6 cases (button, form, dropdown, read table) = 3 secs and 0,02 Cent.", "excerpt": "yes. i used it like that. big model is planning - JEV chooses elements on a site to click. i made a super fast mcp browser. JEV got a list & aim and gives back the probabilities - the tool clicks. result: in 6 cases (button, form, dropdown, read table) = 3 secs and 0,02 Cent. Opus did it in 10 sec vor 11 Cent. crazy imho. didn't do heavy tasks though - pity it can't write text. did anyone had a look at laya yet?", "context": "", "use_case": "choosing page elements to click in browser automation", "switched_from": "", "story": "I was impressed by Jev, please explain why I shouldn't be"}, {"id": 798141521, "item_id": 49883845, "url": "https://news.ycombinator.com/item?id=49883845", "by": "firelex", "date": "2026-09-28", "stance": "compare", "quote": "On multi-step reasoning it's behind: BBH 64-68% against Jev's 94%", "excerpt": "Hi HN. Jeff is a set of small, open-weight Qwen3.5 and Gemma fine-tunes for zero-shot classification, with respectable out-of-the-box performance, meant to be slotted right into code (or fine-tuned further as needed). You give them a situation and a list of options; they return a calibrated probability for each, in one forward pass, with no text generation. The 2B scores 83.1% on a five-benchmark panel (Jev's publish", "context": "", "use_case": "", "switched_from": "", "story": "Jeff \u2013 Jev-compatible 0.8B decision models, trained at home, ~30 ms"}, {"id": 798039073, "item_id": 49877442, "url": "https://news.ycombinator.com/item?id=49877442", "by": "bothlabs", "date": "2026-09-28", "stance": "recommend", "quote": "it performs on par\nwith small LLMs while being faster and cheaper", "excerpt": "I built jevpipe as a wrapper for TypeSafe's Jev System 1 model. Jev answers yes/no, picks from a few choices or gives a score, with probabilities, and it's fast and cheap. See https://docs.typesafe.ai/introduction for more details on Jev. It performs on par with small LLMs while being faster and cheaper (independent study: https://www.ayautomate.com/blog/jev-vs-llm-benchmark), and in my benchmark it was better calibr", "context": "", "use_case": "fast yes/no classification", "switched_from": "", "story": "Show HN: Jevpipe \u2013 a fast System 1 for AI agents, as a Unix pipe"}, {"id": 797667328, "item_id": 49854208, "url": "https://news.ycombinator.com/item?id=49854208", "by": "prathje", "date": "2026-09-26", "stance": "compare", "quote": "Is Jev just that with nice caching?", "excerpt": "Nice! I would love to use it for images as well. Then again is using Grammar-Based Decoding with a json response not the same? Is Jev just that with nice caching? Because then I have been using that already\u2026", "context": "", "use_case": "structured LLM output", "switched_from": "", "story": "A single function Jev-like wrapper for LLMs, including vision models"}, {"id": 797584192, "item_id": 49849012, "url": "https://news.ycombinator.com/item?id=49849012", "by": "shepardrtc", "date": "2026-09-25", "stance": "recommend", "quote": "It really does just work.  And it works so well I already integrated it into my product.", "excerpt": "It really does just work. And it works so well I already integrated it into my product. Saves me about 75% of costs for the section its working in, which isn't a small amount. I see a lot of negativity and I don't really get it either. Its so cheap and so fast, why not give it a try?", "context": "builds a product using it", "use_case": "cost-saving classification in product", "switched_from": "", "story": "Ollaya \u2013 Ollama for open-source, Jev-style decision models"}, {"id": 797524128, "item_id": 49845258, "url": "https://news.ycombinator.com/item?id=49845258", "by": "matt_oriordan", "date": "2026-09-25", "stance": "recommend", "quote": "Jev is a great model, unblocks things you could not do before when you're in builder mode", "excerpt": "Author here, and I'll declare my bias so I'm upfront. I am co-founder of Ably, a realtime infrastructure company, so I wanted Jev to be a low-latency story. Two weeks ago I built Jev Pong and thought it was. Then I tried to find a use our customers would ship and couldn't, so I sorted the whole first week of Jev posts instead of guessing. The data is on a separate page linked from the piece (see https://mattheworiord", "context": "co-founder of Ably", "use_case": "builder-mode prototyping", "switched_from": "", "story": "Jev Is Built for Tinkerers"}, {"id": 797133360, "item_id": 49820835, "url": "https://news.ycombinator.com/item?id=49820835", "by": "visarga", "date": "2026-09-23", "stance": "compare", "quote": "Jev, on the other hand, has a generic model, not trained on in-domain examples, so its confidence scores are uncalibrated for any non-generic task.", "excerpt": "I also built one, but mine uses embeddings. It classifies concepts defined by a collection of positive and negative examples. The classifier model is trained in <1 second using ridge regression. The model itself is exactly the same shape as the embedding, so it works as a concept embedding. Since I already have a dataset, I can use it to do conformal prediction in order to calibrate confidence scores. Jev, on the oth", "context": "built their own embedding-based classifier", "use_case": "text classification", "switched_from": "", "story": "Jev in 25 Lines of Python"}, {"id": 797035888, "item_id": 49814743, "url": "https://news.ycombinator.com/item?id=49814743", "by": "CROON_tv", "date": "2026-09-23", "stance": "recommend", "quote": "It's fast and cheap: about 0.35 s per decision, about 2 cents per hour of audio.", "excerpt": "I use Jev in an iPhone app that dubs whatever the phone plays in real time. It decides whether a translated sentence is complete enough to speak now or should wait. It's fast and cheap: about 0.35 s per decision, about 2 cents per hour of audio.", "context": "built an iPhone dubbing app", "use_case": "real-time audio dubbing", "switched_from": "", "story": "Ask HN: Jev - Anyone built anything useful for day to day work with Jev?"}, {"id": 796999618, "item_id": 49812476, "url": "https://news.ycombinator.com/item?id=49812476", "by": "bladeaidev", "date": "2026-09-23", "stance": "compare", "quote": "Zero-shot Laya tracks Jev on 2-4 labels and falls off a cliff on 77", "excerpt": "Same finding from a different angle. Zero-shot Laya tracks Jev on 2-4 labels and falls off a cliff on 77 (38% vs 76% on banking77 in jevbench). What closed the gap for me was not a bigger model but a small head trained on the task's own rows on top of the frozen encoder: a 12-label intent site went 89.5% -> 100% agreement with its teacher on 3000 rows, support triage 69/34/66% -> 99/86/98%, with the head answering 90", "context": "", "use_case": "intent classification", "switched_from": "", "story": "Does an open-weight decision model beat a hosted one? Jev vs. Laya"}, {"id": 796883648, "item_id": 49805228, "url": "https://news.ycombinator.com/item?id=49805228", "by": "vardalab", "date": "2026-09-22", "stance": "recommend", "quote": "Now that Jev is out I can finally have a true AI sysadmins managing my \"cloud in the basement\" homelab", "excerpt": "Yeah, flash models, DeepSeek, MiMo, GLM, I love those things. For simple tasks like a daily routine shit, just setting up stuff and then doing the hard stuff in Claude/Codex, that's a reasonable approach for someone like me, a \"gentleman code farmer\", lol. And even lower tier stuff, I have the local models taking care of. Now that Jev is out I can finally have a true AI sysadmins managing my \"cloud in the basement\" h", "context": "homelab user", "use_case": "homelab sysadmin", "switched_from": "", "story": "Claude Opus 5.5"}, {"id": 796856449, "item_id": 49803528, "url": "https://news.ycombinator.com/item?id=49803528", "by": "suraj_phanindra", "date": "2026-09-22", "stance": "neutral", "quote": "the absence of essential options can cause jev to choose the wrong option with high confidence", "excerpt": "i built wellposed as a plugin/skill which works with any agent (https://github.com/suraj-phanindra/wellposed) to ensure your agent understands how to choose the right kind of jev request and format it correctly not just for syntax but for completeness and correctness. it is well-documented in the \"jagged-ness\" docs (https://docs.typesafe.ai/model-jaggedness/jev-1.13) that typesafe include on their docs page that the ", "context": "", "use_case": "", "switched_from": "", "story": "Jev \u2013 A curation of Jev demos on X, tools, skills, and integrations"}, {"id": 796700480, "item_id": 49793780, "url": "https://news.ycombinator.com/item?id=49793780", "by": "pampas", "date": "2026-09-21", "stance": "recommend", "quote": "I used Jev to make my AI game 6x faster and 8.7x cheaper.", "excerpt": "I used Jev to make my AI game 6x faster and 8.7x cheaper.", "context": "", "use_case": "AI game", "switched_from": "", "story": "Jev is an honest game changer"}, {"id": 796699776, "item_id": 49793736, "url": "https://news.ycombinator.com/item?id=49793736", "by": "lanyard-textile", "date": "2026-09-21", "stance": "neutral", "quote": "I use my Jev BYOK", "excerpt": "Thank you ;) +1 on the animation. Thanks for the feedback. Fortunately it's on OpenRouter for everyone now :) I think. I use my Jev BYOK, but I assume you don't need the key.", "context": "", "use_case": "trolley problem demo", "switched_from": "", "story": "Show HN: Will Jev pull the lever in the trolley problem?"}, {"id": 796593648, "item_id": 49787103, "url": "https://news.ycombinator.com/item?id=49787103", "by": "ranyume", "date": "2026-09-21", "stance": "warn", "quote": "jev as it is now is really bad at making connections and understanding it's input", "excerpt": "I tried to use jev for this. I'll share what I learned for the interested. -- The setup was a simple map with different rooms. Each room had 1-3 doors. For the inputs: The AI had an array of \"known places\" empty at the start, the current position, the current doors with no information about where they lead to, and the list of past actions The goal / task for the AI was to explore all rooms and save them to known plac", "context": "", "use_case": "NPC decision-making in games", "switched_from": "", "story": "Kev: Tiny Jev-like family of decision models built on top of Qwen3.5"}, {"id": 796578864, "item_id": 49786179, "url": "https://news.ycombinator.com/item?id=49786179", "by": "jdw64", "date": "2026-09-21", "stance": "neutral", "quote": "I use it mainly to check whether this code fits the rules I defined", "excerpt": "JEV seems similar to BERT. Where would it be useful? I use it mainly to check whether this code fits the rules I defined, just a yes or no. But I'm not sure if that's the right way to use it.", "context": "", "use_case": "yes/no rule checking on code", "switched_from": "", "story": "Jev-Leftpad"}, {"id": 796389424, "item_id": 49774339, "url": "https://news.ycombinator.com/item?id=49774339", "by": "kncvetko", "date": "2026-09-20", "stance": "warn", "quote": "but with 40.7% accuracy it's basically useless for this use case.", "excerpt": "Is Jev the general-purpose classifier we\u2019ve been waiting for? That would be the thing that could create a real dent in the universe: a classifier that can classify without first collecting a massive statistical dataset and training it with whatever algorithm fits best; be it BERT or good old logistic regression. Long story short: no. The more nuanced answer is: it depends. Over the years, we\u2019ve built classification s", "context": "built classification systems for credit scoring and tax tech under German regulation", "use_case": "bank transaction classification", "switched_from": "", "story": "Is Jev the general-purpose classifier we've been waiting for?"}, {"id": 796342528, "item_id": 49771408, "url": "https://news.ycombinator.com/item?id=49771408", "by": "calebkaiser", "date": "2026-09-20", "stance": "recommend", "quote": "Jev seems pretty cool! I just got access and have only gotten to do minimal experiments", "excerpt": "Jev seems pretty cool! I just got access and have only gotten to do minimal experiments, but I love this general area of research and it fills a very real need. I agree with you. I think the OPs pushback is emblematic of a larger reaction I've seen that is, at the very least, misinformed. There are a lot of approaches that use a self-attention backbone for classifier-style outputs. You have structured generation libr", "context": "", "use_case": "schema-guided classification", "switched_from": "", "story": "I built non-autoregressive decision models with RL a year ago"}, {"id": 796295937, "item_id": 49768496, "url": "https://news.ycombinator.com/item?id=49768496", "by": "MiguelG719", "date": "2026-09-19", "stance": "recommend", "quote": "act: 4.3x faster, 97% fewer LLM calls. Pass rate: 97.5% -> 98.3%.", "excerpt": "We also just added Jev support https://github.com/browserbase/stagehand/pull/2952 On the act/extract/observe evals it shows promising results being extremely efficient - act: 4.3x faster, 97% fewer LLM calls. Pass rate: 97.5% -> 98.3%. - heldout: 4.1x faster, 78% fewer LLM calls. Pass rate: 87.5% -> 97.5%. - observe: 11.1x faster, 69% fewer LLM calls. Pass rate: 75.0% -> 83.3%. - extract: 8.7x faster, 75% fewer LLM c", "context": "", "use_case": "browser automation evals", "switched_from": "", "story": "We made Playwright 2x faster and 80% more token efficient"}, {"id": 796138704, "item_id": 49758669, "url": "https://news.ycombinator.com/item?id=49758669", "by": "silbercue", "date": "2026-09-18", "stance": "recommend", "quote": "21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good.", "excerpt": "I used it today as browser agent, the \"tool\" is a menu of ~10\u201340 clickable refs from the accessibility tree. Jev picks one per step. 21\u201323 decisions for six benchmark cards, all correct (!), ~$0.001 total. Really good. The part it can't do is write the text for an input field so a nano model does that when Jev picks \"type\". Numbers and code in my top-level comment in this thread. Working very good.", "context": "", "use_case": "browser agent action selection", "switched_from": "", "story": "Introducing System One Models and Jev"}]}