OpenAI Releases GPT-5.6 Series Medical Evaluation Results
- Sophie Larsen
- 4 days ago
- 2 min read
OpenAI released evaluation results for its GPT-5.6 model series on medical tasks.
The tests covered both patient-facing and clinical workflows.
Smaller variants matched or exceeded earlier top models while using far fewer resources.
GPT-5.6 Luna outperformed the strongest GPT-5.5 setup at one-twenty-fifth the inference cost.
GPT-5.6 Sol set the highest scores across the full test suite.
Doctors competed directly against the models
Specialists wrote answers with unlimited time and full web access.
Separate reviewers scored every response without knowing the source.
Scoring covered accuracy, clarity, completeness, instruction following, and usefulness for health decisions.
Twenty thousand individual ratings produced the final rankings.
All GPT-5.6 variants scored above the physician baseline.
Reviewers also identified fewer errors in the model answers than in the human ones.
Smaller model beats prior flagship at lower cost
GPT-5.6 Luna reached its results at minimal reasoning effort.
It still surpassed the best GPT-5.5 configuration that used maximum reasoning.
The cost gap reached twenty-five times.
The larger GPT-5.6 Sol pushed overall performance further while keeping the same efficiency pattern.
These shifts occurred on the same evaluation set used for earlier models.
Evaluation design stressed real clinical conditions
Tasks spanned symptom checking, diagnosis support, treatment planning, and patient communication.
Each doctor answer received blind review from peers.
The five scoring dimensions forced explicit comparison on concrete quality points.
The volume of ratings reduced the chance that single outliers drove the outcome.
Model claims still require external checks
The source of the data is a single company post.
No independent lab has reproduced the twenty-thousand-rating set.
Future studies will need to test the models on fresh patient cases not seen in training.
What to watch next
Track whether other labs run the same tasks on their latest models within the next quarter.
Watch for published peer review of the raw response data.
Monitor regulatory guidance on AI use in documented clinical workflows.
