Microsoft’s BioGPT recorded 45,315 monthly downloads on Hugging Face as of December 2025, making it one of the most widely used domain-specific biomedical language models available. Released in October 2022, BioGPT achieved 78.2% accuracy on the PubMedQA benchmark — a record at the time of publication. This post covers the key BioGPT user data and statistics for 2026, including its community adoption, benchmark performance, and the healthcare AI market surrounding it.
BioGPT Statistics (2026): Key Numbers
- 45,315 monthly downloads recorded on Hugging Face as of December 2025
- 4,500+ GitHub stars and 475 forks in the BioGPT repository as of December 2025
- 78.2% accuracy achieved on the PubMedQA biomedical question-answering benchmark
- 129 BioGPT-tagged models hosted on the Hugging Face platform, including 63 fine-tuned community derivatives
- $613.81 billion — projected value of the global AI healthcare market by 2034, growing at 36.83% CAGR from $26.69 billion in 2024
What Is BioGPT?
BioGPT is a generative Transformer language model developed by Microsoft Research and published in Briefings in Bioinformatics in September 2022. It was built on the GPT-2 architecture and pre-trained entirely on biomedical literature, making it distinct from general-purpose models that require domain adaptation after training.
The base model contains 347 million parameters across 24 Transformer layers, with 1,024 hidden units and 16 attention heads. Microsoft trained it on 15 million PubMed abstracts spanning publications from the 1960s through 2021, using 8 NVIDIA V100 GPUs over 200,000 training steps. A larger variant, BioGPT-Large, extends these capabilities for more demanding downstream tasks.
| Attribute | BioGPT (Base) | BioGPT-Large |
|---|---|---|
| Parameters | 347 million | 1.5 billion |
| Transformer Layers | 24 | — |
| Hidden Units | 1,024 | — |
| Attention Heads | 16 | — |
| Vocabulary Size | 42,384 tokens | 42,384 tokens |
| Training Corpus | 15M PubMed abstracts | 15M PubMed abstracts |
| Training Steps | 200,000 | — |
| PubMedQA Accuracy | 78.2% | 81.0% |
Source: Microsoft Research / Briefings in Bioinformatics (2022), Hugging Face Model Cards
BioGPT Download and Community Adoption Statistics
BioGPT has maintained steady developer interest since its release. The model recorded 45,315 monthly downloads on Hugging Face as of December 2025. The GitHub repository has 4,500+ stars with 475 forks and 74 active watchers monitoring updates. Hugging Face hosts 129 BioGPT-tagged models — 63 of which are fine-tuned community derivatives covering specialized applications from drug discovery to clinical documentation.
Source: Hugging Face Hub / GitHub (December 2025)
The platform also shows 291 likes on the primary model card and 85+ active Spaces built on BioGPT functionality. That level of community activity places BioGPT among the most cited and forked domain-specific NLP models in biomedical research. For context on how cloud services support tools like these, the cloud market share data for 2025 shows AI-specific services growing 140–180% year over year.
BioGPT Benchmark Performance
Microsoft evaluated BioGPT on six biomedical NLP tasks. The model set new records on four of them at the time of publication, outperforming both BioBERT and general-purpose GPT-2 models across relation extraction and question answering.
| Task | Dataset | Metric | BioGPT Score |
|---|---|---|---|
| Chemical-Disease Relation Extraction | BC5CDR | F1 | 44.98% |
| Drug-Target Interaction | KD-DTI | F1 | 38.42% |
| Drug-Drug Interaction | DDI | F1 | 40.76% |
| Question Answering | PubMedQA | Accuracy | 78.2% |
| Question Answering (Large) | PubMedQA | Accuracy | 81.0% |
Source: Luo et al., arXiv:2210.10341 / Briefings in Bioinformatics, 2022
On the BC5CDR relation extraction task specifically, BioGPT outperformed the REBEL model by 8.28 percentage points and bested seq2rel by 4.78 percentage points, despite seq2rel using both training and validation sets. These results established BioGPT as a state-of-the-art model for end-to-end biomedical NLP at launch. Researchers using statistical analysis programs on Chromebook who work in research settings may find BioGPT relevant as a browser-accessible model integration.
Source: Microsoft Research, arXiv:2210.10341 (2022)
BioGPT and the AI Healthcare Market
BioGPT operates within a fast-growing segment. The global AI healthcare market reached $26.69 billion in 2024 and analysts project it will grow to $36.96 billion in 2025. Long-term forecasts place the market at $613.81 billion by 2034, driven by a compound annual growth rate of 36.83%. North America holds 45% of that market, with the United States alone accounting for $8.41 billion in 2024.
Source: Grand View Research / Market Research Reports (2024–2025 projections)
Software solutions — the category that includes language models like BioGPT — hold a 44.60% share of the AI healthcare market. The FDA had approved over 950 AI-based medical devices as of May 2025, indicating broader regulatory acceptance of AI in clinical contexts. Enterprise technology adoption data shows healthcare consistently leading cloud-tool integration across industries.
Physician AI Adoption
Physician use of health AI reached 66% in 2024, up from 38% in 2023 — a 78% increase in a single year. That acceleration reflects growing acceptance of AI-assisted documentation and research support in clinical settings. BioGPT, as a research-grade tool trained on peer-reviewed literature, sits closer to the research end of that pipeline rather than direct clinical deployment.
| Year | Physician AI Adoption Rate | Year-over-Year Change |
|---|---|---|
| 2023 | 38% | — |
| 2024 | 66% | +28 percentage points (+78%) |
Source: AI Healthcare Adoption Survey, 2024
Who Uses BioGPT?
BioGPT’s user base is primarily academic researchers, pharmaceutical developers, and biomedical NLP practitioners. The model supports four main task categories: relation extraction, question answering, document classification, and text generation. Because it runs through the Hugging Face Transformers library, it integrates into Python-based research pipelines with minimal setup. Microsoft distributes seven fine-tuned checkpoints through official download channels, each optimized for a different downstream biomedical task.
The open-source MIT license covers both the model weights and the code, which has contributed to its community adoption. Researchers who use AI platforms for research workflows increasingly rely on models like BioGPT to process and extract information from scientific literature at scale. Separately, Google Workspace integration statistics point to growing reliance on web-native tools in research and education settings — the same environments where BioGPT tends to appear.
The pharmaceutical research and clinical documentation markets are the most active application areas. Drug-target interaction extraction (KD-DTI), chemical-disease mapping (BC5CDR), and drug-drug interaction detection (DDI) directly align with pharmaceutical literature mining workflows. Chromebook usage statistics for 2026 show that cloud-based tools now handle complex workloads that once required local software installations, which makes browser-accessible models increasingly practical for research teams. Institutions running Google for Education tools also represent a growing segment, given biomedical NLP’s expansion into graduate and postgraduate teaching.
FAQ
How many times has BioGPT been downloaded?
BioGPT recorded 45,315 monthly downloads on Hugging Face as of December 2025. The GitHub repository has over 4,500 stars and 475 forks, reflecting sustained developer interest since its 2022 release.
What accuracy does BioGPT achieve on PubMedQA?
The base BioGPT model achieves 78.2% accuracy on PubMedQA. The larger BioGPT-Large variant reaches 81.0% on the same benchmark, both setting records at the time of publication.
How large is the AI healthcare market in 2024?
The global AI healthcare market reached $26.69 billion in 2024. It is projected to grow to $613.81 billion by 2034 at a 36.83% compound annual growth rate, with North America holding 45% of the current market.
What data was BioGPT trained on?
BioGPT was trained on 15 million PubMed abstracts covering publications from the 1960s through 2021. Training ran for 200,000 steps on 8 NVIDIA V100 GPUs using the PyTorch and Hugging Face Transformers frameworks.
Is BioGPT free to use?
Yes. BioGPT is released under the MIT License, which covers both the model code and pre-trained weights. It is available through the Hugging Face Hub and Microsoft’s official GitHub repository at no cost.
