Which model leads HealthBench Consensus length-adjusted?

HealthBench Consensus length-adjusted Leaderboard — Model Scores

Pro Mode — raw scores, one benchmark under the lensofficial card harness

The field · 12 measured

94.496.3

12 models measured, most on the official card harness. GPT-5.4 tops the board at 96.3.

12 measured·Lifecycle Updated Oct 8, 2026Score basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 12Bars show 94.4–96.3 · ◆ solid = first-party or better · outlined = arm's-length

01GPT-5.4OpenAI96.3$2.5/M$15.0/M$0.091

02GPT-6 SolOpenAI96.2$2.0/M$10.0/M$0.062

03GPT-6.1 SolOpenAI96$2.0/M$10.0/M$0.063

04GPT-6 LunaOpenAI95.9$0.10/M$0.50/M$0.003

05GPT-5.5+1 altOpenAI95.6$5.0/M$30.0/M$0.183

06GPT-5OpenAI95.6$1.3/M$10.0/M$0.059

07GPT-5.6 SolOpenAI95.5$4.0/M$20.0/M$0.126

08GPT-6 Astra+1 altOpenAI95.5$10.0/M$50.0/M$0.314

09GPT-5.6 TerraOpenAI95.1$2.0/M$12.0/M$0.074

10GPT-5.6 Luna+1 altOpenAI95.1$0.20/M$1.2/M$0.007

11GPT-5.1OpenAI95$1.3/M$10.0/M$0.059

12GPT-5.2OpenAI94.4$1.8/M$14.0/M$0.083

◆ solid = independently verified · aggregator attested · vendor attributed — outlined = cross-reference · source-attributed. Dimmed ◆ = confidence inferred. “+N alts” = alternate disclosures in the drawer with why-demoted. Prices per million tokens; $/point = input price ÷ score. Every row opens its model dossier.

Verification: 12 models scored · 0 independently verified · 12 vendor-reported · 0 with source disagreement. How these tiers are assigned