# New publicly available MedCAT models

**URL:** <https://discourse.cogstack.org/t/new-publicly-available-medcat-models/373>\
**Category:** MedCAT\
**Created:** [October 10, 2025, 2:57pm UTC](https://discourse.cogstack.org/t/new-publicly-available-medcat-models/373 "2025-10-10T14:57:58Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![mart.ratas](https://avatars.discourse-cdn.com/v4/letter/m/e79b87/32.png) [@mart.ratas](https://discourse.cogstack.org/u/mart.ratas)\
**Post date:** [October 10, 2025, 2:57pm UTC](https://discourse.cogstack.org/t/new-publicly-available-medcat-models/373/1 "2025-10-10T14:57:58Z")

</div>

Hello everyone,

With the semi recent release of MedCAT v2 we figured it’s finally time to release a few newer publicly available Snomed models. These are available through the download link in the [README](https://github.com/CogStack/cogstack-nlp/tree/main/medcat-v2). Or just [here](https://medcat.sites.er.kcl.ac.uk/auth-callback-api) directly with a valid UMLS API key.

We have 2 public **v2** models available:

1. SnomedCT UK Clinical edition 39.0 (Oct 2024) and UK Drug Extension 39.0 (July 2024) based model enriched with UMLS 2024AA; trained only on MIMIC-IV
2. SnomedCT UK Clinical edition 40.2 (June 2025) and UK Drug Extension 40.3 (July 2024) based model enriched with UMLS 2024AA; trained only on MIMIC-IV

We also have a number of MedCAT **v1** models available:

1. UMLS Small (A modelpack containing a subset of UMLS (disorders, symptoms, medications…). Trained on MIMIC-III)
2. SNOMED International (Full SNOMED modelpack trained on MIMIC-III)
3. UMLS Dutch v1.10 (a modelpack provided by UMC Utrecht containing [UMLS entities with Dutch names](https://github.com/umcu/dutch-umls) trained on Dutch medical wikipedia articles and a negation detection model [repository](https://github.com/umcu/negation-detection/)/[paper](https://doi.org/10.48550/arxiv.2209.00470) trained on EMC Dutch Clinical Corpus).
4. UMLS Full. \>4MM concepts trained self-supervised on MIMIC-III. v2022AA of UMLS.
5. The same 2024 based model as above in v1 format
6. The same 2025 based model as above in v1 format

While we encourage you use MedCAT v2 and the models in that native format, if you download an older version MedCAT v2 will be able to load it and covnert it to the format it knows. However, the loading process will be considerably longerin those cases.

EDIT:  
We do not guarantee any performance metrics for the publicly available models.

---

<div class="post-metadata">

**Author:** ![gmag](https://avatars.discourse-cdn.com/v4/letter/g/bbce88/32.png) [@gmag](https://discourse.cogstack.org/u/gmag)\
**Post date:** [October 18, 2025, 1:12pm UTC](https://discourse.cogstack.org/t/new-publicly-available-medcat-models/373/2 "2025-10-18T13:12:07Z")

</div>

Thank you 🙂

For medcatv2 model:

v2 Snomed 2025 MIMIC trained (This is a model based on the 2025 July release enriched with names from UMLS 2024AA. It’s been trained on MIMIC-IV notes.)

Are ICD10, UMLS codes included? I thought from the description they would be? When I run the example below (used in medcatv1 docs with what appears to be just an earlier model, does include ICD10) I don’t seem to get them this time around? Or perhaps I am missing a function argument or cat.config detail?

```auto
text = “The patient was diagnosed with leukemia.”
entities = cat.get_entities(text)
import jsonprint(json.dumps(entities, indent=2))

```

```auto
{“entities”: {“0”: {“pretty_name”: “Patient”,“cui”: “116154003”,“type_ids”: [“31601201”],“source_value”: “patient”,“detected_name”: “patient”,“acc”: 1,“context_similarity”: 1,“start”: 4,“end”: 11,“id”: 0,“meta_anns”: {},“context_left”: ,“context_center”: ,“context_right”: },“1”: {“pretty_name”: “Diagnosis”,“cui”: “439401001”,“type_ids”: [“2680757”],“source_value”: “diagnosed”,“detected_name”: “diagnosed”,“acc”: 1,“context_similarity”: 1,“start”: 16,“end”: 25,“id”: 1,“meta_anns”: {},“context_left”: ,“context_center”: ,“context_right”: },“2”: {“pretty_name”: “Leukemia”,“cui”: “1162768007”,“type_ids”: [“33782986”],“source_value”: “leukemia”,“detected_name”: “leukemia”,“acc”: 0.99,“context_similarity”: 0.99,“start”: 31,“end”: 39,“id”: 2,“meta_anns”: {},“context_left”: ,“context_center”: ,“context_right”: }},“tokens”: }

```

Are these additional arguments for get\_entities I need to flag to pull these? I was getting them in the prior medcatv1 and based on the model detail you provide, I’d have thought they would appear.

Lastly - would the models contain [assertion detection](https://arxiv.org/pdf/2503.17425v1) as well?

---

<div class="post-metadata">

**Author:** ![mart.ratas](https://avatars.discourse-cdn.com/v4/letter/m/e79b87/32.png) [@mart.ratas](https://discourse.cogstack.org/u/mart.ratas)\
**Post date:** [October 20, 2025, 8:40am UTC](https://discourse.cogstack.org/t/new-publicly-available-medcat-models/373/3 "2025-10-20T08:40:11Z")

</div>

Hi! I already answered the ICD-10 mapping question on the issues page, so feel free to refer to that:

> <https://github.com/CogStack/cogstack-nlp/issues/96#issuecomment-3418693875>
>
> I have access to UMLS via NIH but when I click the v1 link in the README I get a…n access denied. Is there any other way to obtain the models? I'd be happy to try and get v2 working but I can't proceed much beyond .load() at the moment :| 
> 
> I see on the docs they are "coming soon" though I notice the hive of activity on the repo - is the release anytime soon? If not, how might I get the v1 models? 
> 
> I just had a few synthetic EHR records I wanted to convert to snowmed or UMLS codes.
> 
> thank you and amazing work

As for assertion detection, MedCAT takes a slightly different approach. We call these Meta-annotations. That is to say, we generally group these into Presence (True, False, or Hypothetical), Subject (Patient, Relative, or Other), and Time (Recent, Past, Future). You can take a [the tutorials](https://github.com/CogStack/cogstack-nlp/tree/main/medcat-v2-tutorials) for _some_ guidance. [The v1 tutorials](https://github.com/CogStack/cogstack-nlp/tree/main/v1/medcat-tutorials) have some more details, but are not directly applicable to v2.

We do have some _very basic_ MetaCAT models floating around _somewhere_. But if you wish to have better performance, you’d probably need to train your own.

PS: There’s no included MetaCAT models shipped with (at least the new) the model packs.