GPT-5.6 Sol · HealthBench(48,000+ rubric items across 5,000 multi-turn conversations) · 附件证据(gpt56-card-healthbench-method)
GPT-5.6 Sol 的 HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations 官方出处。见《GPT-5.6 System Card》§5.1 HealthBench · PDF physical p.15 / printed p.14。证据 gpt56-card-healthbench-method。
原文与定位
GPT-5.6 Sol · HealthBench(48,000+ rubric items across 5,000 multi-turn conversations),GPT-5.6 Sol HealthBench evaluations配置,长度校正分(0–100) 57 score(具体成绩公布时间未确认)。见《GPT-5.6 System Card》§5.1 HealthBench · PDF physical p.15 / printed p.14。 查看成绩条件
GPT-5.6 Sol · HealthBench(48,000+ rubric items across 5,000 multi-turn conversations),GPT-5.6 Sol HealthBench evaluations配置,未校正分(0–100) 55.6 score(具体成绩公布时间未确认)。见《GPT-5.6 System Card》§5.1 HealthBench · PDF physical p.15 / printed p.14。 查看成绩条件
GPT-5.6 Sol · HealthBench(48,000+ rubric items across 5,000 multi-turn conversations),GPT-5.6 Sol HealthBench evaluations配置,Mean response length 1764 characters(具体成绩公布时间未确认)。见《GPT-5.6 System Card》§5.1 HealthBench · PDF physical p.15 / printed p.14。 查看成绩条件
For this system card, we report all variants for completeness. Responses of 2,000 characters receive no adjustment. Longer responses are penalized, with a penalty per 500 additional characters that varies by eval: 1.47 points per 500 characters for HealthBench Professional, 2.99 for HealthBench, 3.92 for HealthBench Hard, and 0.20 for HealthBench Consensus. Shorter responses receive a corresponding positive adjustment. All penalties here are reported on the 0-100 scale that we report this eval on.
- 官方材料
- GPT-5.6 System Card · PDF 第 15 页
- 定位
- 附件 · §5.1 HealthBench · PDF physical p.15 / printed p.14
- 材料版本
- Cover title GPT-5.6 System Card, cover date 2026-07-09; PDF metadata title says GPT-5.6 Preview System Card. Current PDF includes change-log entries dated 2026-08-03 and 2026-08-19; publication date of this current PDF version is not stated.
- 版本公布
- 未确认
- 披露时间
- 未确认 · 首发/后补未确认
- 采集时间
- 历史依据
- 暂无可靠历史版本依据
引用文本
GPT-5.6 Sol · HealthBench(48,000+ rubric items across 5,000 multi-turn conversations),GPT-5.6 Sol HealthBench evaluations配置,长度校正分(0–100) 57 score(具体成绩公布时间未确认);GPT-5.6 Sol · HealthBench(48,000+ rubric items across 5,000 multi-turn conversations),GPT-5.6 Sol HealthBench evaluations配置,未校正分(0–100) 55.6 score(具体成绩公布时间未确认);GPT-5.6 Sol · HealthBench(48,000+ rubric items across 5,000 multi-turn conversations),GPT-5.6 Sol HealthBench evaluations配置,Mean response length 1764 characters(具体成绩公布时间未确认)。官方出处 GPT-5.6 System Card(附件,Cover title GPT-5.6 System Card, cover date 2026-07-09; PDF metadata title says GPT-5.6 Preview System Card. Current PDF includes change-log entries dated 2026-08-03 and 2026-08-19; publication date of this current PDF version is not stated.)§5.1 HealthBench · PDF physical p.15 / printed p.14,https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf#page=15。Benchlog https://benchlog.lekshome.cc/evidence/gpt-5-6/gpt56-card-healthbench-method/
关联成绩与条件
| 指标与成绩 | 配置与评测条件 | 证据关系 |
|---|---|---|
| 长度校正分(0–100)57 score原文 57.0 · 单位 score 公布 未确认 | GPT-5.6 Sol HealthBench evaluations
| 直接支持该成绩 在排名中查看此成绩 |
| 未校正分(0–100)55.6 score原文 55.6 · 单位 score 公布 未确认 | GPT-5.6 Sol HealthBench evaluations
| 直接支持该成绩 在排名中查看此成绩 |
| Mean response length1764 characters原文 1764 · 单位 characters 公布 未确认 | GPT-5.6 Sol HealthBench evaluations
| 直接支持该成绩 在排名中查看此成绩 |