HealthBench Consensus (updated implementation)
Professional Work
HealthBench Consensus rubric score from OpenAI's updated HealthBench implementation, reported unadjusted in the August and October 2026 ChatGPT system cards. This correlated HealthBench metric is diagnostic; the cards do not establish its exact item count. Preserve the raw value separately from the response-length adjustment.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Luna (ChatGPT, October 2026) | 95.5 |
| GPT-5.6 Sol (ChatGPT, August 2026) | 95.3 |
| GPT-5.6 Luna (ChatGPT, August 2026) | 94.6 |
| GPT-6 Sol (ChatGPT, October 2026) | 94.0 |