--- title: JEV vs BERT: What’s Actually Different? | daily.dev description: A comparison between JEV, a new closed-source classification API from TypeSafe AI, and classic BERT/ModernBERT classifiers. JEV requires no fine-tuning, lets... canonical: https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: JEV vs BERT: What’s Actually Different? | daily.dev og:description: A comparison between JEV, a new closed-source classification API from TypeSafe AI, and classic BERT/ModernBERT classifiers. JEV requires no fine-tuning, lets... og:url: https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr og:image: https://api.daily.dev/og/posts/9ngCESlfr.png og:image:alt: JEV vs BERT: What’s Actually Different? og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # JEV vs BERT: What’s Actually Different? **[Medium](https://daily.dev/sources/medium_js)** · 6 min read · 3 upvotes · 0 comments ## Summary A comparison between JEV, a new closed-source classification API from TypeSafe AI, and classic BERT/ModernBERT classifiers. JEV requires no fine-tuning, lets you specify labels per API call (up to 255), supports three answer types (choice, noul, score), and claims calibrated confidence scores via a proprietary method called RLCD. Benchmarks from Sebastian Raschka show JEV and fine-tuned ModernBERT perform similarly on IMDb sentiment classification (96.5% vs 95% accuracy), with JEV actually slower than an already-trained BERT model despite marketing claims of being 20-200x faster (that comparison was against token-by-token chatbots, not BERT). The architecture is likely a small encoder similar to ModernBERT, with the real differentiator being JEV's claimed 100% synthetic training data. Trade-offs: BERT is self-hostable, private, and fast once trained; JEV is closed, flexible across tasks, but requires API calls and trusting an unpublished calibration method. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does JEV compare to a fine-tuned ModernBERT classifier for sentiment analysis accuracy and speed? On a 25,000-review IMDb sentiment test by Sebastian Raschka, fine-tuned ModernBERT reached about 95% accuracy and classified all reviews in roughly 7 minutes once trained (after a 23-minute fine-tuning step), costing compute for training. JEV reached about 96.5% accuracy with zero training, but took about 22 minutes through the API for roughly 65 cents, meaning JEV was not faster than an already-trained BERT model. _Weighing BERT versus JEV for a classification task? daily.dev surfaces benchmarks like this for that exact decision._ ### What is RLCD in the JEV classification model? RLCD stands for Reinforcement Learning for Calibrated Decisions, a term coined by TypeSafe AI, the company behind JEV, to describe the training method meant to make the model's confidence scores trustworthy (so an 0.8 score should be correct about 8 out of 10 times). It is not an established academic technique, and the company has not yet published calibration results to verify the claim. _Anyone relying on model confidence scores for automation can track claims like this via daily.dev._ ### How many labels can a single JEV API call classify text into? A single JEV call can include up to 255 label options, sent as part of the request rather than baked into the model architecture like a traditional BERT classifier. This is possible because JEV reportedly uses one shared scoring layer that scores every option and applies softmax, letting the label set change freely from call to call without retraining. _Developers building flexible text routing or triage systems can follow API capabilities like this through daily.dev._ --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm), [#nlp](https://daily.dev/tags/nlp), [#bert](https://daily.dev/tags/bert) [View this post on daily.dev](https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"JEV vs BERT: What’s Actually Different?","url":"https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr"},"datePublished":"2026-10-02T20:00:21.274Z","dateModified":"2026-10-02T20:20:20.742Z","description":"A comparison between JEV, a new closed-source classification API from TypeSafe AI, and classic BERT/ModernBERT classifiers. JEV requires no fine-tuning, lets...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c6951a5e76e33c8a64fb7e1a7e990ddb?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c6951a5e76e33c8a64fb7e1a7e990ddb?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,llm,nlp,bert","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"JEV vs BERT: What’s Actually Different?"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr#faq","mainEntity":[{"@type":"Question","name":"How does JEV compare to a fine-tuned ModernBERT classifier for sentiment analysis accuracy and speed?","acceptedAnswer":{"@type":"Answer","text":"On a 25,000-review IMDb sentiment test by Sebastian Raschka, fine-tuned ModernBERT reached about 95% accuracy and classified all reviews in roughly 7 minutes once trained (after a 23-minute fine-tuning step), costing compute for training. JEV reached about 96.5% accuracy with zero training, but took about 22 minutes through the API for roughly 65 cents, meaning JEV was not faster than an already-trained BERT model. Weighing BERT versus JEV for a classification task? daily.dev surfaces benchmarks like this for that exact decision."}},{"@type":"Question","name":"What is RLCD in the JEV classification model?","acceptedAnswer":{"@type":"Answer","text":"RLCD stands for Reinforcement Learning for Calibrated Decisions, a term coined by TypeSafe AI, the company behind JEV, to describe the training method meant to make the model's confidence scores trustworthy (so an 0.8 score should be correct about 8 out of 10 times). It is not an established academic technique, and the company has not yet published calibration results to verify the claim. Anyone relying on model confidence scores for automation can track claims like this via daily.dev."}},{"@type":"Question","name":"How many labels can a single JEV API call classify text into?","acceptedAnswer":{"@type":"Answer","text":"A single JEV call can include up to 255 label options, sent as part of the request rather than baked into the model architecture like a traditional BERT classifier. This is possible because JEV reportedly uses one shared scoring layer that scores every option and applies softmax, letting the label set change freely from call to call without retraining. Developers building flexible text routing or triage systems can follow API capabilities like this through daily.dev."}}]} ```