| <!doctype html> |
| <html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><title>OpenJev · Decisions for your product</title><link rel="stylesheet" href="style.css"></head><body> |
| <div class="openjev-card" style="font-family:Inter,-apple-system,BlinkMacSystemFont,Arial,sans-serif;color:inherit;line-height:1.6;max-width:1100px;margin:auto"> |
| <p style="font-size:12px;font-weight:700;letter-spacing:2px;text-transform:uppercase;color:light-dark(#247653,#65dba4);margin:0 0 14px">OpenJev · Open weights for real products</p> |
| <p style="font-size:18px;line-height:1.65;max-width:820px;margin:0 0 24px">Give OpenJev text or a screenshot and your options. Get back a decision and a score.</p> |
| <p style="font-size:15px;line-height:1.7;max-width:880px;margin:0 0 24px">A decision model has to be small enough and fast enough to sit inside a product, and still be general: any situation, any set of choices, given at request time. Small task-specific models are fast, but they are not general. OpenJev is a 27B model that answers in one pass on a single GPU and comes close to Jev: 84.0% against 85.4% on our 10,000-question text benchmark. OpenJev Flash 9B is the smaller one: about 1.6x faster, 79.4% on the same 10,000 questions, and on par with Cloudflare's Clef-Flash and ahead of Kev-9B and Nimble 9B on JevBench.</p> |
| <a href="https://www.runpod.io/blog/how-to-deploy-openjev-on-runpod-serverless" target="_blank" rel="noopener noreferrer" style="display:block;border:1px solid light-dark(#247653,#65dba4);background:light-dark(#2476530f,#65dba41a);border-radius:10px;padding:14px 18px;margin:0 0 24px;color:inherit;text-decoration:none"><strong style="color:light-dark(#247653,#65dba4)">New: deploy OpenJev on Runpod Serverless.</strong> Runpod published a step-by-step guide for running OpenJev on one H100, no GPU of your own needed. <strong>Read the guide ↗</strong></a> |
| <div style="display:flex;flex-wrap:wrap;gap:12px;margin:0 0 30px"> |
| <a href="https://huggingface.co/openjev/openjev" target="_blank" rel="noopener noreferrer" style="display:inline-block;background:#247653;color:white;text-decoration:none;font-weight:700;padding:11px 20px;border-radius:8px">Get OpenJev →</a> |
| <a href="https://www.runpod.io/blog/how-to-deploy-openjev-on-runpod-serverless" target="_blank" rel="noopener noreferrer" style="display:inline-block;border:1px solid #7f948b;color:inherit;text-decoration:none;font-weight:700;padding:10px 20px;border-radius:8px">Deploy on Runpod ↗</a> |
| <a href="https://www.loopai.com/" target="_blank" rel="noopener noreferrer" style="display:inline-block;border:1px solid #7f948b;color:inherit;text-decoration:none;font-weight:700;padding:10px 20px;border-radius:8px">Build with Loop ↗</a> |
| </div> |
| <div style="display:flex;flex-wrap:wrap;gap:24px;border-top:1px solid #82948955;border-bottom:1px solid #82948955;padding:22px 0;margin:0 0 10px"> |
| <div style="flex:1 1 180px"><strong style="display:block;font-size:34px;line-height:1.2;color:light-dark(#247653,#65dba4)">84.0%</strong><span>Text decision accuracy</span><br><small style="opacity:.72">10,000 questions · main model</small></div> |
| <div style="flex:1 1 180px"><strong style="display:block;font-size:34px;line-height:1.2;color:light-dark(#247653,#65dba4)">88.0%</strong><span>Desktop screenshot decisions</span><br><small style="opacity:.72">2,000 next-action steps · main model</small></div> |
| <div style="flex:1 1 180px"><strong style="display:block;font-size:34px;line-height:1.2;color:light-dark(#247653,#65dba4)">~15 GB</strong><span>MLX 4-bit model</span><br><small style="opacity:.72">Run text decisions on Apple silicon</small></div> |
| <div style="flex:1 1 180px"><strong style="display:block;font-size:34px;line-height:1.2;color:light-dark(#247653,#65dba4)">~40 ms</strong><span>OpenJev Flash 9B</span><br><small style="opacity:.72">Per decision on one H100 · 1.6x faster</small></div> |
| </div> |
| <p style="font-size:12px;opacity:.8;margin:0 0 26px"><a href="https://huggingface.co/openjev/openjev#results" target="_blank" rel="noopener noreferrer">Explore the benchmarks and methods →</a></p> |
| <h2 style="font-size:21px;letter-spacing:-.4px;margin:0 0 14px">One model, useful across your business.</h2> |
| <div style="display:flex;flex-wrap:wrap;gap:24px;margin:0 0 28px"> |
| <div style="flex:1 1 190px"><strong>Customer operations</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Sort requests, route tickets and flag urgent messages using your own rules.</p></div> |
| <div style="flex:1 1 190px"><strong>Billing and documents</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Classify invoices, apply approval criteria and send exceptions for review.</p></div> |
| <div style="flex:1 1 190px"><strong>Agents and QA</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Score allowed actions from page text or screenshots to power your browser workflows.</p></div> |
| </div> |
| <h2 style="font-size:21px;letter-spacing:-.4px;margin:0 0 12px">Choose where you run it.</h2> |
| <div style="display:flex;flex-wrap:wrap;gap:10px;margin:0 0 18px"> |
| <a href="https://huggingface.co/openjev/openjev" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev ↗</strong><br><small style="opacity:.75">BF16 · text and screenshots</small></a> |
| <a href="https://huggingface.co/openjev/openjev-FP8" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev FP8 ↗</strong><br><small style="opacity:.75">GPU · text and screenshots</small></a> |
| <a href="https://huggingface.co/openjev/openjev-MLX" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev MLX ↗</strong><br><small style="opacity:.75">8-bit · Apple silicon · text</small></a> |
| <a href="https://huggingface.co/openjev/openjev-MLX-4bit" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev MLX 4-bit ↗</strong><br><small style="opacity:.75">About 15 GB · Apple silicon · text</small></a> |
| <a href="https://huggingface.co/openjev/OpenJev-Flash-9B" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev Flash 9B ↗</strong><br><small style="opacity:.75">9B · BF16 · text, faster</small></a> |
| <a href="https://huggingface.co/openjev/OpenJev-Flash-9B-MLX" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev Flash 9B MLX ↗</strong><br><small style="opacity:.75">9B · 8-bit · Apple silicon</small></a> |
| <a href="https://huggingface.co/openjev/OpenJev-Flash-9B-MLX-4bit" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev Flash 9B MLX 4-bit ↗</strong><br><small style="opacity:.75">9B · about 5 GB · Apple silicon</small></a> |
| <a href="https://huggingface.co/openjev/OpenJev-Flash-9B-FP8" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev Flash 9B FP8 ↗</strong><br><small style="opacity:.75">9B · 11.9 GB · GPU</small></a> |
| <a href="https://huggingface.co/openjev/OpenJev-Flash-9B-GGUF" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>OpenJev Flash 9B GGUF ↗</strong><br><small style="opacity:.75">9B · llama.cpp · from 5.6 GB</small></a> |
| </div> |
| <p style="font-size:13px;opacity:.8;margin:0 0 18px">Your labels are supplied at request time. One forward pass per question with up to 52 choices. Open weights for research and non-commercial use under CC BY-NC 4.0. Commercial licence: <a href="mailto:support@loopai.com">support@loopai.com</a>.</p> |
| <h2 style="font-size:21px;letter-spacing:-.4px;margin:18px 0 8px">One server. Any model.</h2> |
| <p style="margin:0 0 12px;font-size:15px">Run the same decision API on your own hardware, with OpenJev or any open model you choose. Send a situation and your questions, get one choice, yes/no probability or score per question, with a probability for every option. Fits a new model to honest probabilities with one command. GPU or Mac. <code>pip install</code> and one command to serve.</p> |
| <div style="display:flex;flex-wrap:wrap;gap:10px;margin:0 0 18px"> |
| <a href="https://huggingface.co/spaces/openjev/openjev-server" target="_blank" rel="noopener noreferrer" style="flex:1 1 175px;border:1px solid #82948966;border-radius:9px;padding:14px;color:inherit;text-decoration:none"><strong>openjev-server ↗</strong><br><small style="opacity:.75">The decision API as a server · Apache 2.0</small></a> |
| </div> |
| <h2 style="font-size:21px;letter-spacing:-.4px;margin:18px 0 8px">For enterprises.</h2> |
| <p style="margin:0 0 14px;font-size:15px"><strong>Policies change. The model does not need to.</strong> The rules go in the request; the labels come back with probabilities, one pass each. Measured on generated cases with exact ground truth and one public dataset. <a href="https://github.com/loopai-hq/openjev-server/blob/master/demos/README.md" target="_blank" rel="noopener noreferrer">See the demos →</a></p> |
| <div style="display:flex;flex-wrap:wrap;gap:12px;margin:0 0 22px"> |
| <div style="flex:1 1 200px;border:1px solid #82948966;border-radius:9px;padding:14px"><strong style="display:block;font-size:30px;line-height:1.2;color:light-dark(#247653,#65dba4)">93.5%</strong><span>policy decisions right</span><br><small style="opacity:.72">200 cases, nine-rule returns policy; 88.0% when it counts calendar days itself; majority baseline 42%</small></div> |
| <div style="flex:1 1 200px;border:1px solid #82948966;border-radius:9px;padding:14px"><strong style="display:block;font-size:30px;line-height:1.2;color:light-dark(#247653,#65dba4)">200 of 200</strong><span>agent tool calls gated correctly</span><br><small style="opacity:.72">seven-rule tool policy; 0 of 84 forbidden calls executed</small></div> |
| <div style="flex:1 1 200px;border:1px solid #82948966;border-radius:9px;padding:14px"><strong style="display:block;font-size:30px;line-height:1.2;color:light-dark(#247653,#65dba4)">95.0%</strong><span>prompt injections caught</span><br><small style="opacity:.72">662 public messages; 97.7% recall at a 2% false-alarm rate</small></div> |
| </div> |
| <div style="display:flex;flex-wrap:wrap;gap:24px;margin:0 0 28px"> |
| <div style="flex:1 1 190px"><strong>Runs where your data is</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Your GPU, your VPC or a Mac. Nothing leaves your network.</p></div> |
| <div style="flex:1 1 190px"><strong>Probabilities you can act on</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Calibrated scores: auto-approve above a threshold, send the rest to a person.</p></div> |
| <div style="flex:1 1 190px"><strong>One API, any model</strong><p style="margin:5px 0;font-size:14px;opacity:.85"><a href="https://huggingface.co/spaces/openjev/openjev-server" target="_blank" rel="noopener noreferrer">openjev-server</a> serves OpenJev or any open model behind the same interface.</p></div> |
| <div style="flex:1 1 190px"><strong>Commercial use</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Weights are CC BY-NC 4.0. For a commercial licence, email <a href="mailto:support@loopai.com">support@loopai.com</a>.</p></div> |
| <div style="flex:1 1 190px"><strong>Text and screens</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Text, JSON, DOM or a screenshot, one endpoint. 88.0% on 2,000 desktop next-action steps.</p></div> |
| <div style="flex:1 1 190px"><strong>Many questions, one call</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Route, score urgency and flag review in the same request; questions share the input but never see each other's answer.</p></div> |
| <div style="flex:1 1 190px"><strong>Fast enough for the request path</strong><p style="margin:5px 0;font-size:14px;opacity:.85">About 125 ms per decision on one H100, about 150 ms on a Mac, up to 52 options in one pass.</p></div> |
| <div style="flex:1 1 190px"><strong>No lock-in</strong><p style="margin:5px 0;font-size:14px;opacity:.85">Open weights you can audit, and the same API in front of any open model, so the vendor is never the bottleneck.</p></div> |
| </div> |
| <p style="margin:0 0 28px;font-size:15px"><strong>Why trust it.</strong> Most small "decision models" only score well on their makers' own tests. On the same 10,000 public questions, same options, same order for every model: hosted Jev 85.4%, OpenJev 84.0%, the fine-tuned 9B we tested 75.7%. On par with the hosted model, eight points ahead of the small ones, and the method is public, so you can rerun it.</p> |
| <p style="font-size:12px;opacity:.65;margin:20px 0 0;padding-top:14px;border-top:1px solid #82948944">OpenJev is an independent project, not affiliated with TypeSafe; Jev is their product.</p> |
| </div> |
| </body></html> |
|
|
