Abacus.AI 推出智能体长程基准 LiveBench 2.0
精选理由
Abacus.AI 做了个反刷分的长程智能体基准 LiveBench 2.0，还会算每任务成本，等 Argon 发布时一起上线，选模型时可以盯着它看。
Abacus.AI 发布新基准 LiveBench 2.0，定位为智能体长程任务评测，将随其 Argon 模型一同上线。该基准强调反映真实世界模型表现，声称无法被刷分（benchmaxx），并计入每任务成本。目前已有数百个基准但大多失效，LiveBench 2.0 针对此问题设计，长程任务类别是其重点。
原文 · Bindu Reddy
Super Excited For Our New Agentic Long Horizon Benchmark - LiveBench 2.0
We have hundreds of benchmarks, but most of them are just plain broken and bench-maxxed. We have worked over the last several weeks to fix this
- launching whenever Argon launches - reflects real-world model performance - you can't benchmaxx on it - long-horizon category is spot on - accounts for cost/task
The first step towards legit SuperIntelligence is for benchmarks to reflect real-world performance