The Builder

Open-weight vision model fine-tuned for radiology and clinical imaging

“The primitive here is a vision-language model with a domain-specific instruction fine-tune released as open weights — that's a real, nameable thing, and it matters. The DX bet is correct: drop the weights on Hugging Face under a research license so a team can pull them with one `transformers` call and run inference on-prem, which is exactly what hospital IT requires. The moment of truth is the first inference call with a DICOM-converted PNG — if the system prompt examples in the model card are solid, this survives the 10-minute test; if they're vague, researchers are on their own. My one gripe: the research license creates a hard fork from the permissive Llama community, so every downstream fine-tune has to re-negotiate terms, and that friction is a real DX tax.”

Free AI workspace for verified US physicians — GPT-5.4, clinical search, and CME credits

“The reusable skills feature for clinical workflows is the killer feature here — automating prior auth paperwork alone could save hours per week per clinician. And the HealthBench score outperforming human physicians given unlimited time is a genuine benchmark result, not a cherry-picked marketing number. OpenAI built something substantial.”