Raw, unadjusted score. The report gives the length-adjusted score over all effort levels only as a 64.7%-65.4% range; per-effort exact chart values were not recovered.
Sampling settings not stated unless recorded per result.
工具
工具条件未记录
推理预算
low
框架
未记录
协议
HealthBench length-adjusted score.
条件
Figure 8.15.B labels the shared Low-to-High result as 64.9%; this row uses the Low effort config. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.
Sampling settings not stated unless recorded per result.
工具
工具条件未记录
推理预算
medium
框架
未记录
协议
HealthBench length-adjusted score.
条件
Figure 8.15.B labels the shared Low-to-High result as 64.9%; this row uses the Medium effort config. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.
Sampling settings not stated unless recorded per result.
工具
工具条件未记录
推理预算
high
框架
未记录
协议
HealthBench length-adjusted score.
条件
Figure 8.15.B labels the shared Low-to-High result as 64.9%; this row uses the High effort config. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.
Sampling settings not stated unless recorded per result.
工具
工具条件未记录
推理预算
xhigh
框架
未记录
协议
HealthBench length-adjusted score.
条件
Figure labels the effort Xhigh. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.
Claude Sonnet · HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations · Release history
View scores, configurations, and disclosures for HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations across 2 Claude Sonnet releases, ordered by release date.
Differences are calculated only for unique, active, exact values between direct predecessors when the unit, metric, and explicitly recorded configuration and conditions match.
· 5 score records for this release, including 5 exact values (may include historical records)
Records for the direct predecessor do not provide matching conditions for a valid difference.
Claude Sonnet 5.5 · HealthBench scores and configurations
Metric and score
Configuration and evaluation conditions
Sources
Raw score69.4%Reported 69.4% · Unit %
Published Unconfirmed
Max effort
Settings
Max reasoning effort.
Unknowns
Adaptive thinking status is benchmark-specific.
Tools
Tool access not recorded
Reasoning budget
max
Framework
not recorded
Protocol
HealthBench, max effort.
Conditions
Raw, unadjusted score. The report gives the length-adjusted score over all effort levels only as a 64.7%-65.4% range; per-effort exact chart values were not recovered.
Sampling settings not stated unless recorded per result.
Tools
Tool access not recorded
Reasoning budget
low
Framework
not recorded
Protocol
HealthBench length-adjusted score.
Conditions
Figure 8.15.B labels the shared Low-to-High result as 64.9%; this row uses the Low effort config. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.
Sampling settings not stated unless recorded per result.
Tools
Tool access not recorded
Reasoning budget
medium
Framework
not recorded
Protocol
HealthBench length-adjusted score.
Conditions
Figure 8.15.B labels the shared Low-to-High result as 64.9%; this row uses the Medium effort config. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.
Sampling settings not stated unless recorded per result.
Tools
Tool access not recorded
Reasoning budget
high
Framework
not recorded
Protocol
HealthBench length-adjusted score.
Conditions
Figure 8.15.B labels the shared Low-to-High result as 64.9%; this row uses the High effort config. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.
Sampling settings not stated unless recorded per result.
Tools
Tool access not recorded
Reasoning budget
xhigh
Framework
not recorded
Protocol
HealthBench length-adjusted score.
Conditions
Figure labels the effort Xhigh. Solid scores reflect length adjustment; all models used safety classifiers and Claude Opus 4.8 graded HealthBench responses.