Event ·Multi-source

医疗长文本基准对比:Anthropic 回答更全,OpenAI 更准更简洁

Artificial Analysis 在 MLCR-AA 和 HealthBench Professional 两个医疗基准上对比了 14 个模型。Claude Sonnet 5.5、Opus 5.5 和 Fable 5.1 占据 MLCR-AA 完整性前三,比最好的 OpenAI 模型高出约 19-23 分,Opus 和 Sonnet 在 HealthBench Professional 原始分上也以 71.4% 和 70.8% 领先 GPT-6 Astra 的 69.7%。但 OpenAI 模型在准确性上占优:Anthropic 模型每个回答的篇幅约为 OpenAI 的 4-6 倍,准确性得分只有 19-61%,OpenAI 则为 89-97%。在同时奖励两者的综合分上,GPT-6 Astra 与 Claude Sonnet 5.5 并列 MLCR-AA 第一,并在 HealthBench Professional 长度调整分上小幅领先。

Current read

Artificial Analysis 拿 14 个模型在两个医疗基准上掰头,结论挺反直觉:Claude 写得多反而更容易出错,GPT-6 综合分才最均衡。

11 sources58° AI heatLast updated 2026/10/22 02:04:56

Evidence

11 sources
Primary sourceArtificial Analysis
View source
Related source 1TestingCatalog
View source
Related source 2ARC Prize
View source
Related source 4Ethan Mollick
View source
Related source 6arXiv: DeepSeek
View source
Related source 7Tanishq Abraham (论文推介)
View source
Related source 8swyx (AI Engineer)
View source
Related source 9rohanpaul_ai
View source
Related source 10@cb_doge
View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.