MedErtas-v0.1-31B-LS
Built on Gemma 4 31B
Best for
- Medical knowledge and codes
- Answer checking
- Spotting unanswerable questions
- Output speed
- 189 tokens/s
- Weights
- 24.2 GB, NVFP4
- Drafter
- Perigee™
- Licence
- Apache-2.0
MedErtas is a family of open-weight language models for medical AI companies, built for fast, low-cost serving.
Faster than baseline.
Accuracy over MedGemma 27B.
Per million input tokens.
Built on Gemma 4 31B
Best for
Built on Qwen3.8-27B
Best for
| Task | 31B LS | 27B LS | Leads | MedGemma 27Breference |
|---|---|---|---|---|
| Clinical calculationMedCalc-Bench 1.2 | 63.9 | 77.3 | 27B LS | 48.6 |
| Calculation with toolsMedCalc with tools | 61.1 | 76.3 | 27B LS | NA |
| Medical knowledgeMedQA | 93.0 | 86.7 | 31B LS | 87.4 |
| Unanswerable questionsMetaMedQA | 70.4 | 65.4 | 31B LS | 36.2 |
| Medical codesMedConceptsQA, mean of medium and hard | 69.8 | 63.2 | 31B LS | 46.4 |
| Answer checkingMedHallu, mean of 3 levels | 76.7 | 72.3 | 31B LS | 64.7 |
| Long patient recordsLongHealth, mean of 2 tasks | 90.0 | 89.0 | Either27B costs less on long inputs | 81.5 |
Developers can adapt MedErtas to clinical workflows and run it privately within agentic systems.
Answer questions over long patient histories.
See exampleExtract variables and work through clinical scores.
See exampleReview medical answers for unsupported claims.
See exampleExplain radiology report terms in plain language.
See exampleAn 87-year-old man was admitted to our hospital for anorexia for several days, high-grade fever from the previous day, and liver dysfunction. Of note, he had a history of hypertension, diabetes mellitus (DM), and angina. Physical examination findings included: clear consciousness; height, 163 cm; weight, 48 kg; blood pressure, 66/40 mmHg; heart rate, 75/min; respiratory rate, 22/min; oxygen saturation of peripheral artery, 96%; and body temperature, 38.1 °C. He had no surface lymphadenopathy. Laboratory findings included: white blood cell (WBC) count, 4.2 × 109/L; hemoglobin, 9.6 g/dL; platelet count, 106 × 109/L; lactate dehydrogenase (LDH), 1662 IU/L; aspartate aminotransferase (AST), 6562 IU/L; alanine aminotransferase (ALT), 1407 IU/L; alkaline phosphatase (ALP), 509 IU/L; γ-glutamyl transpeptidase (γ-GTP), 130 IU/L; total bilirubin, 2.7 mg/dL; prothrombin time (PT), 20.3 s; international normalized ratio (INR), 1.73; blood urea nitrogen (BUN), 34.4 mg/dL; and creatinine, 1.4 mg/dL. Serologic tests were all negative for hepatitis A, B, C, and E. Computed tomography (CT) performed at admission revealed no distinct nodules or masses in the liver, but mild splenomegaly, multiple hypodense splenic lesions, and para-aortic lymph node swelling (Fig. a–c). Antibiotics were administered intravenously for suspected acute cholangitis and sepsis, but the fever persisted. Abdominal ultrasonography (US) and magnetic resonance imaging (MRI) performed 2 days later newly depicted gallbladder wall thickening and ascites (Figs. , ). On the same night, the patient presented with grade II hepatic encephalopathy, with blood tests showing a decrease in platelet count and worsening liver function suggestive of “acute type” ALF with hepatic coma and disseminated intravascular coagulation (DIC). While consideration was being given to treating acute liver failure, the patient suddenly developed ventricular fibrillation and died despite immediate resuscitation measures on day 3 of hospitalization. Given the rapid clinical course, an autopsy was performed in this patient with suspected ALF likely due to drug-induced liver injury, autoimmune hepatitis, hepatitis of indeterminate etiology, circulatory disturbance, metabolic disease, or malignant cells including malignant lymphoma. This led to the patient being diagnosed with malignant, diffuse large B-cell lymphoma, characterized by multiple lymph-node, as well as multi-organ (i.e., spleen, liver, lung, and prostate) involvement: (1) general swelling of lymph nodes in the para-abdominal aorta (Fig. a), peripancreatic, peribronchial, pulmonary hilar, subclavicular regions, as well as in the pelvis, especially around the iliac arteries, measuring up to 60 mm in diameter; (2) presence of multiple whitish splenic nodules (Fig. b) with the entire spleen shown to be microscopically infiltrated by tumor cells (Fig. a); (3) irregular geographical morphology of the liver (Fig. c) with the lymphoma cells mainly infiltrating the portal areas (Fig. b); (4) involvement of the right lobe of the lung; (5) involvement of both lobes of the prostate consequent to outer connective tissue involvement (Fig. c). Microscopically, the tumor cells in the lymph nodes were shown to have medium-sized irregular nuclei and small nucleoli with a high nuclear cytoplasmic ratio, thus showing diffuse proliferation characterized by the presence of large numbers of small T-cells (Fig. ). Immunohistochemistry revealed the tumor cells to be positive for CD20 (Fig. a), CD79a (Fig. b), bcl-2 (Fig. c), and MUM1 (Fig. d), but negative for CD3 (Fig. e), CD10, bcl-6, and cyclin D1, with a Ki-67 labeling index (Ki-67 LI) of 16% (Fig. f). The causes of death were thus determined as (1) DIC and (2) ALF associated with diffuse large B-cell lymphoma.
What is the patient's Creatinine Clearance using the Cockroft-Gault Equation in terms of mL/min?
I'll calculate the Creatinine Clearance using the Cockcroft-Gault equation.
Step 1: Determine BMI and weight category
Let me extract the values from the patient note: - Age: 87 years - Sex: Male - Height: 163 cm - Weight: 48 kg - Creatinine: 1.4 mg/dL
48 / (1.63 * 1.63)BMI ≈ 18.07, which is underweight (BMI < 18.5).
Per the instructions: "If the patient is underweight, please set their adjusted body weight to their actual body weight."
So adjusted body weight = 48 kg (actual weight).
Step 2: Apply Cockcroft-Gault equation for males:
CrCl = [(140 − age) × weight] / (72 × serum creatinine)
((140 - 87) * 48) / (72 * 1.4)Calculation: - (140 − 87) = 53 - 53 × 48 = 2544 - 72 × 1.4 = 100.8 - 2544 / 100.8 ≈ 25.24 mL/min
Recorded prompt, with temperature from the evaluation protocol. Confirm the model identifier and request options supported by your API deployment. Your application must execute the calculator tool loop.
{
"model": "MedErtas-v0.1-27B-LS",
"messages": [
{
"role": "system",
"content": "Please reason step by step, use tools as needed, then give your final answer within <answer>...</answer> XML tags."
},
{
"role": "user",
"content": "You are a helpful assistant. Your task is to read a patient note and compute a medical value based on the following question.\nYou have access to a calculator tool and a python code execution tool. Use them to do any math based on values in the patient note. If using the python tool, make sure the code returns or prints all of its outputs. If there are any errors with compiling or running your script, you may need to re-write your code to obtain the output.\nNote that all of the necessary information is provided in the patient note and you should not need to prompt the user for any information.\nWhen you are finished with all computations, please write your final answer value without any units, using the following formats inside the indicated answer tag:\n\n- Decimal Answer Format: 17.29\n- Score-Based Answer Format: 5\n- Estimated Date Answer Format: 5/21/2021\n- Estimated Age Answer Format: (4 weeks, 3 days)\n\nPatient Note:\n\nAn 87-year-old man was admitted to our hospital for anorexia for several days, high-grade fever from the previous day, and liver dysfunction. Of note, he had a history of hypertension, diabetes mellitus (DM), and angina. Physical examination findings included: clear consciousness; height, 163 cm; weight, 48 kg; blood pressure, 66/40 mmHg; heart rate, 75/min; respiratory rate, 22/min; oxygen saturation of peripheral artery, 96%; and body temperature, 38.1 °C. He had no surface lymphadenopathy. Laboratory findings included: white blood cell (WBC) count, 4.2 × 109/L; hemoglobin, 9.6 g/dL; platelet count, 106 × 109/L; lactate dehydrogenase (LDH), 1662 IU/L; aspartate aminotransferase (AST), 6562 IU/L; alanine aminotransferase (ALT), 1407 IU/L; alkaline phosphatase (ALP), 509 IU/L; γ-glutamyl transpeptidase (γ-GTP), 130 IU/L; total bilirubin, 2.7 mg/dL; prothrombin time (PT), 20.3 s; international normalized ratio (INR), 1.73; blood urea nitrogen (BUN), 34.4 mg/dL; and creatinine, 1.4 mg/dL. Serologic tests were all negative for hepatitis A, B, C, and E. Computed tomography (CT) performed at admission revealed no distinct nodules or masses in the liver, but mild splenomegaly, multiple hypodense splenic lesions, and para-aortic lymph node swelling (Fig. a–c). Antibiotics were administered intravenously for suspected acute cholangitis and sepsis, but the fever persisted. Abdominal ultrasonography (US) and magnetic resonance imaging (MRI) performed 2 days later newly depicted gallbladder wall thickening and ascites (Figs. , ). On the same night, the patient presented with grade II hepatic encephalopathy, with blood tests showing a decrease in platelet count and worsening liver function suggestive of “acute type” ALF with hepatic coma and disseminated intravascular coagulation (DIC). While consideration was being given to treating acute liver failure, the patient suddenly developed ventricular fibrillation and died despite immediate resuscitation measures on day 3 of hospitalization.\nGiven the rapid clinical course, an autopsy was performed in this patient with suspected ALF likely due to drug-induced liver injury, autoimmune hepatitis, hepatitis of indeterminate etiology, circulatory disturbance, metabolic disease, or malignant cells including malignant lymphoma. This led to the patient being diagnosed with malignant, diffuse large B-cell lymphoma, characterized by multiple lymph-node, as well as multi-organ (i.e., spleen, liver, lung, and prostate) involvement: (1) general swelling of lymph nodes in the para-abdominal aorta (Fig. a), peripancreatic, peribronchial, pulmonary hilar, subclavicular regions, as well as in the pelvis, especially around the iliac arteries, measuring up to 60 mm in diameter; (2) presence of multiple whitish splenic nodules (Fig. b) with the entire spleen shown to be microscopically infiltrated by tumor cells (Fig. a); (3) irregular geographical morphology of the liver (Fig. c) with the lymphoma cells mainly infiltrating the portal areas (Fig. b); (4) involvement of the right lobe of the lung; (5) involvement of both lobes of the prostate consequent to outer connective tissue involvement (Fig. c). Microscopically, the tumor cells in the lymph nodes were shown to have medium-sized irregular nuclei and small nucleoli with a high nuclear cytoplasmic ratio, thus showing diffuse proliferation characterized by the presence of large numbers of small T-cells (Fig. ). Immunohistochemistry revealed the tumor cells to be positive for CD20 (Fig. a), CD79a (Fig. b), bcl-2 (Fig. c), and MUM1 (Fig. d), but negative for CD3 (Fig. e), CD10, bcl-6, and cyclin D1, with a Ki-67 labeling index (Ki-67 LI) of 16% (Fig. f). The causes of death were thus determined as (1) DIC and (2) ALF associated with diffuse large B-cell lymphoma.\n\nQuestion:\n\nWhat is the patient's Creatinine Clearance using the Cockroft-Gault Equation in terms of mL/min? You should use the patient's adjusted body weight in kg instead of the patient's actual body weight if the patient is overweight or obese based on their BMI. If the patient's BMI's normal, set their adjusted body weight to the minimum of the ideal body and actual weight. If the patient is underweight, please set their adjusted body weight to their actual body weight. You should use the patient's medical values and health status when they were first admitted to the hospital prior to any treatment."
}
],
"temperature": 1,
"chat_template_kwargs": {
"enable_thinking": false
},
"tools": [
{
"type": "function",
"function": {
"name": "calculator",
"description": "Evaluate a mathematical expression safely.\n\nSupports basic arithmetic (+, -, *, /, %, **), parentheses, and common math functions. Use ^ or ** for exponentiation.\n\nAvailable functions: sqrt, sin, cos, tan, log, log10, exp, abs, round\nAvailable constants: pi, e",
"parameters": {
"properties": {
"expression": {
"description": "A mathematical expression to evaluate.",
"title": "Expression",
"type": "string"
}
},
"required": [
"expression"
],
"title": "calculator_args",
"type": "object",
"additionalProperties": false
},
"strict": null
}
},
{
"type": "function",
"function": {
"name": "code_interpreter",
"description": "Execute Python code in a sandboxed REPL environment. Variables persist across calls.\n\nThis is a restricted environment - file I/O and network access are disabled.\nAllowed imports: math, numpy, scipy. Pre-loaded: numpy (as np), math, scipy. Use print() to display results.",
"parameters": {
"properties": {
"code": {
"description": "A block of Python code to execute.",
"title": "Code",
"type": "string"
}
},
"required": [
"code"
],
"title": "code_interpreter_args",
"type": "object",
"additionalProperties": false
},
"strict": null
}
}
]
}Initiated R-Pola-CHP regimen q21
Which of the following is NOT part of the current treatment regimen for Mrs. Sample's DLBCL?
The correct answer is D: Vincristine.
Recorded prompt, with temperature from the evaluation protocol. Confirm the model identifier and request options supported by your API deployment.
{
"note": "Open patient record or copy the request to load the complete original input.",
"model": "MedErtas-v0.1-31B-LS",
"temperature": 1
}Double balloon enteroscopy: is it efficacious and safe in a community setting?
DBE shows slightly higher complication rates in the community setting due to less experienced staff performing the procedure.
The provided answer makes a generalization that double balloon enteroscopy (DBE) has "slightly higher complication rates in the community setting due to less experienced staff." In medical literature, studies comparing DBE in community settings versus tertiary academic centers generally indicate that DBE is safe and efficacious in both environments, with complication rates remaining low and comparable across settings when performed by trained endoscopists. There is no established medical consensus or consistent evidence that community settings inherently have higher complication rates due to staff experience.
Therefore, the answer provides a claim that is not supported by established clinical evidence and attributes a cause (staff experience) that contradicts the general finding that the procedure is safe in community settings.
Recorded prompt, with temperature from the evaluation protocol. Confirm the model identifier and request options supported by your API deployment.
{
"model": "MedErtas-v0.1-31B-LS",
"messages": [
{
"role": "system",
"content": "\nYou are an AI assistant with extensive knowledge in the field of medicine. I want you act as an answer judge. Given a question and an answer, your objective is to determine if the provided answer contains non-factual or hallucinated information. You SHOULD give your judgement based on the following hallucination types and the world knowledge.\n\nType: Misinterpretation of #Question#: These are hallucinated answers that misunderstands the question, leading to an off-topic or irrelevant response.\nType: Incomplete Information: These are hallucinated answers that Point out what is not true without providing correct information.\nType: Mechanism and Pathway Misattribution - These are hallucinated answer that falsely attribution of biological mechanisms, molecular pathways, or disease processes that contradicts established medical knowledge\nType: Methodological and Evidence Fabrication - Inventing false research methods, statistical data, or specific clinical outcomes\n\n\nYou are trying to determine if there is a factual contradiction between the answer and the world knowledge.\nSome information in the answer might be fabricated. Answer \\boxed{{0}} if the answer is factual and \\boxed{{1}} if the answer is hallucinated and \\boxed{{2}} if you are unsure about the factuality and want to pass. If you are unsure about the answer choose \\boxed{{2}} instead of guessing and making a mistake.\n"
},
{
"role": "user",
"content": "\nQuestion: Double balloon enteroscopy: is it efficacious and safe in a community setting?\nAnswer: DBE shows slightly higher complication rates in the community setting due to less experienced staff performing the procedure.\n\nReturn just the answer, Answer \\boxed{0} if the answer is factual and \\boxed{1} if the answer is hallucinated and \\boxed{2} if you are unsure about the answer and want to pass.\nIf you are unsure about the answer choose \\boxed{2} instead of guessing and making a mistake.\nYour Judgement:\n"
}
],
"temperature": 1,
"chat_template_kwargs": {
"enable_thinking": false
}
}Indication: Persistent cough. No prior imaging available for comparison. Technique: Non-contrast CT of the chest with multiplanar reformats. Findings: A solitary 5 mm solid nodule is present in the peripheral right upper lobe. No focal air-space consolidation. No pleural effusion or pneumothorax. Central airways are patent. No enlarged mediastinal lymph nodes by CT size criteria. No acute osseous abnormality in the imaged chest. Impression: Solitary 5 mm right upper lobe pulmonary nodule. No prior study is available to assess stability. No acute air-space disease.
Highlighted terms and their explanations from MedErtas-v0.1-31B-LS will appear here once the recorded run has been reviewed.
The same prompt goes to both models. At full precision with no drafter, Qwen3.8-27B takes 5.54 seconds for the 144-token answer, at 26 tokens/s. MedErtas-v0.1-27B-LS with the Perigee drafter takes 0.70 seconds, at 207 tokens/s.
| MedErtas v0.1 31B LS | MedErtas v0.1 27B LS | |
|---|---|---|
| Baseline | 22.5 tokens/s | 26 tokens/s |
| Med-calibrated weights | 51 tokens/s2.3x | 60 tokens/s2.3x |
| Speculative drafter | 168 tokens/s3.3x | 188 tokens/s3.1x |
| Perigee™ medical drafter | 189 tokens/s+13%8.4x the baseline | 207 tokens/s+10%8x the baseline |
| Model | Input | Cache read | Output |
|---|---|---|---|
| MedErtas-v0.1-31B-LS | $0.20 | $0.05 | $0.60 |
| MedErtas-v0.1-27B-LS | $0.10 | $0.05 | $1.00 |
MedErtas is a family of open-weight medical language models from Ertas AI. The v0.1 release builds on Gemma 4 31B and Qwen3.8-27B, combining domain-aware 4-bit quantisation with Perigee, our medical drafters, for faster serving.
Start with MedErtas-v0.1-27B-LS for clinical calculation and long-record applications with large inputs. Start with MedErtas-v0.1-31B-LS for medical knowledge, code knowledge and answer checking. The 27B has a lower input price; the 31B has a lower output price. Evaluate accuracy, latency and total token usage on your own workload.
The model weights and the Perigee medical drafters are released under Apache-2.0 on Hugging Face. You can download them and run your own deployment. The hosted API is an optional way to access the models.
On the 11 medical benchmarks the Medmarks leaderboard shares with our runs, MedErtas 31B and 27B average 79.6 and 77.1, 16.0 and 13.5 points above the 63.6 for MedGemma 27B in the leaderboard snapshot of 26 October 2026. These are benchmark results, not evidence of clinical suitability. MedErtas v0.1 is text only and built for fast serving, with domain-aware quantisation and Perigee medical drafters; MedGemma also supports medical image understanding. MedErtas uses Apache-2.0, and MedGemma uses the Health AI Developer Foundations terms. Ertas offers a beta API priced per token, while Google's Vertex AI deployment uses a dedicated endpoint. Check current provider pricing for your deployment.
Yes. The weights and Perigee drafters are downloadable. The weights use NVFP4, which runs on NVIDIA Blackwell GPUs. The 31B model's weights are 24.2 GB, plus the 1.0 GB Perigee drafter; the 27B model's weights are 22 GB. Runtime memory depends on your configuration. Talk to us about your own cloud, on-premise or air-gapped deployment.
Every medical model, including one at full precision, should be built into a carefully engineered and reviewed system before production use. MedErtas aims to match its full-precision base on medical capability while making the model faster and cheaper to serve, and each release publishes the comparison with full precision for every set we test. Compressed builds from other publishers can still lose accuracy on particular medical tasks. Test any model against full precision on your own clinical tasks before relying on it.
LS, for Lightweight Study, is the build profile. An LS build compresses the weights with domain-aware quantisation and pairs them with a drafter trained on the domain, with the model's own training unchanged. Later profiles will cover models post-trained on domain data.
Perigee is the name of the MedErtas medical drafters. A drafter is a small companion model that proposes several tokens ahead for speculative decoding; the main model checks them in one pass and keeps only the tokens it would produce itself, so answers follow the main model and arrive faster. MedErtas-v0.1-31B-LS-Perigee is Google's Gemma 4 31B assistant drafter, fine-tuned on medical answers. MedErtas-v0.1-27B-LS-Perigee is Red Hat's DSpark drafter for Qwen3.8-27B, trained further on medical answers. Both are released under Apache-2.0 on Hugging Face and run with vLLM. The name comes from the point in an orbit closest to the body it circles: the more closely a drafter tracks the main model, the more of its tokens are accepted.
Not yet for intensive or stable production use. The API is in beta while we finish the production offering. The weights are on Hugging Face if you want to run them on your own hardware, and if you're planning production volumes, talk to us.
Two LS builds from Gemma 4 31B and Qwen3.8-27B, each with its Perigee medical drafter, open weights and a hosted beta API.
MedErtas is a starting point for developers building healthcare applications on medical text. Each use case needs its own validation, adaptation or meaningful modification by the developer before the model is relied on.
Outputs from MedErtas models are not intended to directly inform clinical diagnosis, patient management decisions, treatment recommendations or any other direct clinical practice. Benchmark results show baseline capability on the tests we name, and outputs can be inaccurate. Treat every output as preliminary: it needs independent verification and clinical review.