Legora 发布 10 月 BAR 基准结果：Opus 5.5 总榜领先
做法律 AI 或想选模型的看下，Legora 每月用真实律所案例测 17 个模型，还加了按任务类型筛选，比看总榜实用多了。
法律 AI 公司 Legora 发布了 10 月的 BAR（Benchmark for Agentic Reasoning）结果，新增按执业领域、任务类型和模型三重筛选功能，覆盖 30 个执业领域、7 个任务类型和 17 个模型。本月 Claude Opus 5.5 在整体质量上领先，超出平均水平 6%。但总榜排名并不能说明全部：GPT-6.1 Sol 整体仅排第 13 名，却在比较类任务上进入前四。BAR 的测试任务来自真实法律案例，用于追踪前沿模型在法律工作上的实际表现。
The October Legora BAR results are here. Alongside this month’s update, you can now filter results by practice area, task family, and model to explore performance across 30 practice areas, seven task families, and 17 models.
Since launching the Benchmark for Agentic Reasoning, we’ve continued to test the latest frontier models on tasks drawn from real-world cases, tracking what improves legal work as models and our harness evolve.
This month, Opus 5.5 leads on overall quality at 6% above average. But overall rankings don’t tell the whole story: GPT-6.1 Sol ranks 13th overall yet places in the top four on comparison tasks. Filter by task family to see how models perform on the type of work you need done.
Explore all results and the updated BAR: https://t.co/6S38KGssuK