Home Model series GPT unsuffixed models GPT-5.5 GPT-5.5 · HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations · Scores and configurations GPT-5.5 · HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations · Scores and configurations GPT-5.5 record for HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations. 4 score records for this release, including 4 exact values (may include historical records). Includes 1 source location.
Scores and evaluation conditions GPT-5.5 · HealthBench scores and configurations Metric and score Configuration and evaluation conditions Sources 长度校正分(0–100) 56.5 score Reported 56.5 · Unit score Published Unconfirmed
GPT-5.5 model-level safety evaluation Settings GPT-5.5 model results for the production release; each benchmark has its stated prompt, task, and safeguard conditions. Unknowns The card does not state one universal system prompt, reasoning effort, or sampling configuration across evaluations. Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol OpenAI 系统卡 HealthBench 计分;2,000 字符不调整,每多 500 字符减 2.99 分,较短回复对应加分;grader 未说明。 Conditions Length-adjusted score; updated HealthBench implementation. 系统卡明确采用 0–100 量表;保留原始分数字符串,不乘除数值。 未校正分(0–100) 58.4 score Reported 58.4 · Unit score Published Unconfirmed
GPT-5.5 model-level safety evaluation Settings GPT-5.5 model results for the production release; each benchmark has its stated prompt, task, and safeguard conditions. Unknowns The card does not state one universal system prompt, reasoning effort, or sampling configuration across evaluations. Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol OpenAI 系统卡 HealthBench 计分;2,000 字符不调整,每多 500 字符减 2.99 分,较短回复对应加分;grader 未说明。 Conditions Unadjusted score. 系统卡明确采用 0–100 量表;保留原始分数字符串,不乘除数值。 Mean response length 2313 characters Reported 2313 · Unit characters Published Unconfirmed
GPT-5.5 model-level safety evaluation Settings GPT-5.5 model results for the production release; each benchmark has its stated prompt, task, and safeguard conditions. Unknowns The card does not state one universal system prompt, reasoning effort, or sampling configuration across evaluations. Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol OpenAI 系统卡 HealthBench 计分;2,000 字符不调整,每多 500 字符减 2.99 分,较短回复对应加分;grader 未说明。 Conditions Mean response length for HealthBench. 相对 GPT-5.4 的长度校正分差 2.5 points Reported +2.5 · Unit points Published Unconfirmed
GPT-5.5 model-level safety evaluation Settings GPT-5.5 model results for the production release; each benchmark has its stated prompt, task, and safeguard conditions. Unknowns The card does not state one universal system prompt, reasoning effort, or sampling configuration across evaluations. Tools Tool access not recorded Reasoning budget not recorded Framework not recorded Protocol OpenAI 系统卡 HealthBench 计分;2,000 字符不调整,每多 500 字符减 2.99 分,较短回复对应加分;grader 未说明。 Conditions Change in length-adjusted HealthBench score versus GPT-5.4.
Benchmark versions and metrics Aliases HealthBench
Length-adjusted score · % · Higher is better Raw score · % · Higher is better Mean response length · characters · Ranking direction unconfirmed 长度校正分(0–100) · score · Higher is better 未校正分(0–100) · score · Higher is better 相对 GPT-5.4 的长度校正分差 · points · Ranking direction unconfirmed Unadjusted score · % · Higher is better