APEX: General Practitioner (MD) measures how well frontier AI models perform on real primary care physician tasks. Across sample tasks graded by expert physicians, Opus 4.6 Max leads at 71.6% Mean Score, ahead of Opus 4.6 High (70.6%) and Sonnet 4.6 (66.7%).
The table below shows the top 10 AI model productivity scores across 100 General Practitioner (MD) tasks. Scores are captured as of June 2026.
| Rank | Model | Mean Score |
|---|---|---|
| 1 | Opus 4.6 (Max) | 71.6% |
| 2 | Opus 4.6 (High) | 70.6% |
| 3 | Sonnet 4.6 (High) | 66.7% |
| 4 | GPT 5.4 (High) | 64.0% |
| 5 | Gemini 3.1 Pro (High) | 62.7% |
| 6 | GPT 5 (High) | 62.4% |
| 7 | GPT 5.2 (High) | 61.9% |
| 8 | Opus 4.5 (High) | 61.5% |
| 9 | Grok 4 | 60.7% |
| 10 | GPT 5.2 Codex (High) | 59.7% |
View full APEX: General Practitioner (MD) leaderboard →
Who should use APEX: General Practitioner (MD)?
AI productivity index (APEX) leaderboard for General Practitioner (MD) tasks serves 3 types of users:
- Individual physicians and clinicians evaluating which AI model handles clinical reasoning, diagnosis support, and medication management tasks most reliably.
- Health system technology and clinical informatics teams standardizing AI model deployments across clinical services. The benchmark gives them expert-graded performance data on the reasoning tasks physicians actually do.
- Healthcare AI companies building products for clinical decision support that need expert-graded performance data on real primary care tasks.
How are AI model productivity rankings evaluated for general practitioners?
These rankings come from APEX: General Practitioner (MD), Mercor's benchmark for how frontier AI models perform on real primary care work. Unlike general medical knowledge benchmarks that test isolated multiple-choice questions, APEX: General Practitioner (MD) evaluates models on the multi-step clinical reasoning that defines primary care, including medication dosing, drug-drug interactions, and integrating patient history into a management plan.
What does the APEX: General Practitioner (MD) benchmark measure?
APEX: General Practitioner (MD) measures how well frontier AI models perform on primary care tasks across 100 prompts. Each task is passed to models with relevant patient documents, mirroring how physicians actually work. Tasks are graded by human-authored rubrics and scored as Mean Score, which awards partial credit for each criterion a model satisfies.
The benchmark was built by practitioners from University of Pennsylvania, Northwestern, Cornell, Brigham & Women's, and Mount Sinai, and is advised by Eric Topol, cardiologist, geneticist, and founder of the Scripps Research Translational Institute, and a leading voice in digital and precision medicine.
How does APEX: General Practitioner (MD) scoring work?
Every rubric is written by an experienced physician who defines what a correct or complete clinical output looks like for that task. Unlike a binary pass/fail, Mean Score awards credit for each individual rubric criterion a model satisfies. A model that correctly identifies the weight-based dosing rule but misses a relevant drug interaction earns partial credit reflecting the value of what it got right.
This scoring method reflects how clinical reasoning is assessed: identifying the right data points and rules is valuable, but in patient care a missed interaction can change the safe answer entirely, which is why partial credit does not mean partial safety.
What type of task is evaluated by the APEX: General Practitioner (MD) benchmark?
Here's one of the sample tasks that was evaluated within this domain:
Sample task: medication dose adjustment for a patient with a drug interaction
A patient presents with stable liver tests, and the model must determine what changes, if any, should be made to the patient's resmetirom dose, with reasoning, using the attached patient record. Correct handling requires recognizing that the patient's weight drives the base dose, that the patient is also taking loratadine, that loratadine is a moderate CYP2C8 inhibitor, that resmetirom is metabolized by CYP2C8, and that the combination therefore requires a reduced dose.
The rubric has 14 criteria. On this sample task, the top model satisfied 6 of the 14: it correctly identified the signs of possible weight gain, that actual body weight should be measured, and the weight-based dosing threshold. It missed the full drug-drug interaction chain: that loratadine is a CYP2C8 inhibitor, that resmetirom is metabolized by CYP2C8, that the interaction raises resmetirom levels, and the resulting reduced doses and the recommendation to discuss continuing loratadine with the patient.
What are the hardest general practitioner tasks for AI models?
Multi-step clinical reasoning that integrates medication rules with drug-drug interactions remains the hardest category of task on APEX: General Practitioner (MD). Opus 4.6 Max leads at 71.6% Mean Score. The bottom of the ranked models, GLM 5 Thinking at 47.8% and GLM 4.7 at 46.7%, shows that a 25-point spread exists between the strongest and weakest models currently evaluated.
3 types of reasoning consistently separate stronger models from weaker ones, based on the sample task:
- Drug-drug interaction recognition. Identifying that a concurrent medication is an enzyme inhibitor and that the primary drug is metabolized by that same enzyme requires connecting 2 facts that models often handle in isolation but fail to link.
- Interaction-adjusted dosing. Once an interaction is recognized, adjusting the dose correctly depends on it, so a model that misses the interaction produces a confident but unsafe dose recommendation.
- Shared decision-making. Recommending that the clinician discuss the risks and benefits of continuing the interacting medication with the patient reflects the judgment layer of primary care that models frequently omit.
Licensed physicians still outperform AI models on integrated clinical management. The judgment required to catch a hidden interaction, weigh it against the patient's full picture, and decide with the patient remains the preserve of human expertise. AI outputs in clinical settings require physician review before they inform care.
Do AI models or AI tools for clinical work matter more?
The underlying model like Opus 4.6, Sonnet 4.6, or GPT 5.4 determines the quality of clinical reasoning. An AI tool for medical professionals, like an electronic health record assistant, a clinical decision support system, or a documentation tool wraps that capability in context, gives the model access to the patient record and drug databases, and delivers output into the clinical workflow.
A well-integrated AI tool cannot compensate for a weak underlying model on tasks requiring interaction-aware clinical reasoning. Better EHR access does not fix a model's failure to recognize a CYP2C8 interaction. However, a good tool amplifies what the underlying model offers, particularly by surfacing the medication list, lab values, and interaction databases that models need to reason safely.
For patient education content and documentation drafting, the difference between the top models narrows and your tool and workflow often decide the result.
For the interaction-aware clinical reasoning APEX measures, integrating medication rules with drug-drug interactions, the underlying model's capability is the deciding factor. Pick the AI model for the ceiling you need on your most complex primary healthcare cases; pick the tool for how effectively it integrates into your clinical workflow. In all cases, physician oversight remains essential.
Get the full APEX: General Practitioner (MD) dataset
The public leaderboard shows the scores. The full dataset is what health system informatics teams and healthcare AI companies use to make real deployment decisions, including task-level breakdowns and rubric details across all 100 tasks.
For teams that need more
Evaluate AI model performance on your own clinical workflows with custom evaluations graded by Mercor's network of physicians.
Get in touchFrequently Asked Questions
What is the best AI model for general practitioners right now?+−
As of June 2026, Opus 4.6 Max leads APEX: General Practitioner (MD) at 71.6% Mean Score, ahead of Opus 4.6 High (70.6%) and Sonnet 4.6 (66.7%). The top 3 models all come from the same family. Rankings shift with each new model release.
Will AI replace general practitioners?+−
The top model on a benchmark of real primary care tasks satisfies 71.6% of rubric criteria on average across 100 tasks. On the sample dosing task, even the top model satisfies only 6 of 14 criteria, correctly handling weight-based dosing but missing the drug interaction that changes the safe answer. AI supports clinical reasoning, documentation, and patient education. It does not replace the licensed physician judgment that safe patient care requires, and clinical AI outputs require physician review.
Is APEX: General Practitioner (MD) a substitute for clinical validation or regulatory clearance?+−
No. APEX: General Practitioner (MD) is a research benchmark measuring how well models reason on primary care tasks. It is not a medical device, not a substitute for clinical validation, and not regulatory clearance. Any deployment of AI in patient care requires appropriate clinical validation, regulatory review, and physician oversight.
How often is the APEX model productivity leaderboard for General Practitioner (MD) updated?+−
The leaderboard is updated as new frontier models are evaluated. Rankings shift with each major release. Verify scores at the time of your decision on the live leaderboard.
How can my organization evaluate AI models on our own clinical work?+−
Mercor offers custom evaluations graded by vetted physicians on specialty-specific clinical tasks, giving health system informatics teams a benchmark built on their own care settings. Get in touch.
What AI tools do general practitioners use?+−
Ambient documentation tools that draft clinical notes from patient visits are the most widely adopted, built on frontier models and integrated with major EHR systems. Clinical decision support tools and drug-interaction checkers are increasingly incorporating frontier model reasoning. Many clinicians also use Claude or ChatGPT for literature synthesis and patient education drafting, subject to health system data-handling and privacy policies.
Can I use ChatGPT or Claude on protected health information?+−
Before using any AI model on protected health information, confirm the specific data-handling terms for your tier, verify a Business Associate Agreement is in place where HIPAA applies, check your organization's information security and privacy policies, and confirm the deployment meets applicable regulatory requirements for clinical use.