[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-german-law-benchmark-puts-12-llms-to-the-legal-test":10,"sections":40},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":30,"tags":31,"sources":35,"feedback":39,"feedback_at":22,"cost_usd":39,"total_tokens":39},3169,"german-law-benchmark-puts-12-llms-to-the-legal-test","German Law Benchmark Puts 12 LLMs to the Legal Test","A new dataset of 1,127 German legal reasoning tasks finds closed flagship models at the top and AI co-authoring beating solo human work.","A new academic benchmark tests LLM systems on German legal reasoning — and the results favor the big closed models.\n\nBenGER (Benchmark for German Law) combines 596 exam-style case tasks drawn from multiple levels of legal education with 531 shorter doctrinal reasoning tasks, totaling 1,127 problems. Researchers evaluated 12 systems spanning closed flagship, efficiency-oriented, and open-weight categories. Grading used a rubric-aligned LLM-as-a-Judge method cross-validated against a three-blind-human-reviewer pool — six judge families total. The LLM judge correlated with human graders at Pearson r=0.76 and Cohen's kappa=0.60, and two judges from independent providers cleared the Calderon single-reviewer replacement bar, meaning they performed as reliably as a single qualified human reviewer on human-authored solutions.\n\nClosed flagship models led across all three corpora, which will not surprise anyone who has watched these leaderboards. More interesting: human-AI co-creation measurably outperformed unaided human work on the same tasks, adding a data point to the ongoing debate about whether AI is a lawyer's tool or a lawyer's replacement.\n\nGerman legal reasoning is a useful stress test because subsumption — applying a general rule to specific facts in structured steps — is explicit and gradeable in ways that common-law argumentation often is not. Whether the 12 unnamed systems here include the models your firm is already using is, inconveniently, left as an exercise for the reader.","[\"ai\",\"legal\",\"benchmarks\",\"large-language-models\"]","2026-07-01T04:00:00.000Z","2026-07-01T08:52:30.769Z","2026-07-01T08:52:33.665Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"The dek promises '1,100-plus' tasks but the body never gives a combined total — it lists 596 and 531 separately, leaving the headline figure unverified in the text; additionally, the article names no specific models despite testing 12 systems, missing the concrete specifics criterion, and the body omits the 'Calderon single-reviewer replacement bar' detail that would explain what the correlation figures actually mean.","resolved","ai",[30,32,33,34],"legal","benchmarks","large-language-models",[36],{"name":37,"url":38},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.28183",0,{"sections":41},[42,46,51,56,61,66,71,76,81,86,91,95,100,105],{"name":43,"slug":30,"count":44,"latest_published_at":45},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":60},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":62,"slug":63,"count":64,"latest_published_at":65},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":92,"slug":93,"count":89,"latest_published_at":94},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":101,"slug":102,"count":103,"latest_published_at":104},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":106,"slug":107,"count":108,"latest_published_at":109},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]