Karan Singhal
Key Views & Dialogues
Should You Trust ChatGPT With Your Health?
GPT-6 Luna, OpenAI’s cheapest frontier API model, beat 5.6 Sol on HealthBench Professional at 34 times lower cost. Singhal sees a path to commoditized health intelligence, but says it has not yet affected the $5.3T US spend; savings require workflow integration. Epic, UCSF, and AdventHealth offer deployment evidence, including roughly 80% lower administrative workflow costs reported by AdventHealth.
View summary (7)
- OpenAI says more than 300 million people a week use ChatGPT for health questions — the vast majority on the free tier, and most outside the US. Karan Singhal frames the health push as the “make upside happen” leg of the AGI mission: “raise the floor of human health” with “trustworthy, abundant, reliable health intelligence for free or nearly free,” and raise the ceiling with capabilities that didn’t exist before.
- The cost curve is the tradeable claim: GPT-6 Luna, OpenAI’s cheapest frontier API model, beat 5.6 Sol — “literally our best model a month ago” — on HealthBench Professional “despite being thirty-four times cheaper.” Singhal’s explicit bet is “commoditization of health intelligence” (“we don’t wanna hoard that health intelligence”), while conceding health AI “has not yet had a macro effect” on the $5.3T US spend — compression requires integration and workflow deployment, not just cheap tokens.
- Singhal makes the case for specialized training inside a general model, based on his experience building specialized models: a healthcare-only model “would fall behind the frontier relatively quickly.” OpenAI adds specialized healthcare data into each phase of training across its ChatGPT models shipped over the past year — “specialized training in a general model… the best of both worlds,” “hitching a ride on that scaling frontier.”
- Trust is built through expert aggregation: “if you ask five physicians a question, you’ll get, like, eight different answers,” so OpenAI has a global network of hundreds of physicians across dozens of specialties and countries. When user comfort conflicts with clinical reality, “it’s basically a prioritization of safety” — the model must say “you should seek care” even when the user wants reassurance. The system-level opportunity: clinical guidelines can take “five to 17 years” to penetrate practice, and point-of-care AI could help compress that lag to “weeks or days or even hours.”
- OpenAI says AI can’t provide all of medical care: “I don’t think AI can or will provide all of that.” The strategy is to make “the last mile of actually having impact on patients just look like the last hundred meters for all of our partners” — ChatGPT as “a quarterback for your health,” not a replacement for care providers. His advice to healthcare builders: “bet on the models getting better. That’s probably the most important thing.”
- Enterprise traction is concrete: Epic integration, UCSF using it for chart review, and AdventHealth reporting “roughly eighty percent reduction” in administrative workflow costs. On agents, Singhal argues healthcare’s legacy stack — EHRs running on VMs, no “MCP connector tomorrow” — is ripe for computer-use agents, predicting traditional barriers “will dissolve in the coming months.” Rughani cites a New York Times article and a Blue Cross Blue Shield Association analysis, saying billing and paperwork bots increased spend by “a billion dollars or something like that”; Singhal calls the arms race “a bit of a short-term distraction.”
- Singhal’s biggest-impact five-to-ten-year prediction is new treatments from AI-guided target and molecule generation plus AI-guided prioritization and validation of clinical trials, where candidate molecules already outnumber trial slots. Nearer term, he’s looking for a “move thirty-seven for health” — predictions from longitudinal data that may seem brilliant only in hindsight, because “a human can’t integrate all that data in their head.”
Universal Medical Intelligence: OpenAI’s Plan to Elevate Human Health, with Karan Singhal
OpenAI’s health strategy is scaling through ChatGPT Health, which reaches more than 230 million weekly users and is planned to connect records, wearables, and Apple Health while remaining free without rate limits and excluding health data from foundation-model training. Frontier models reached attending-physician territory in one hospitalization, yet embodied context remains a human edge; HealthBench Hard is only around 40%, while Penda Health’s randomized study shows statistically significant outcome improvements and adoption is targeted for the end of 2026.
View summary (7)
OpenAI is using health as a major way to make its mission real and deliver benefits at mass scale, with more than 230 million people using ChatGPT for health and wellness queries weekly. ChatGPT Health is planned to connect medical records, Apple Health, and wearables while remaining free without rate limits for all users; Karan Singhal says ads are not planned “right now,” creating an early version of what Nathan Labenz calls “universal basic intelligence.”
Labenz reports that frontier models reached attending-physician territory in his son’s case, while the remaining human edge belonged largely to embodied clinical context. HealthBench Hard rose from 0% for GPT-4o when created to roughly 40% for current OpenAI models, versus the “20 range” for current competitors. During his son’s hospitalization, Labenz found frontier models “step for step with the attending oncologist,” but doctors retained an advantage from directly observing breathing, color, and overall appearance.
OpenAI’s health program is being built around expert feedback, evaluation infrastructure, and workflow evidence rather than connected patient-data training. More than 250 physicians—about 260, by Singhal’s estimate—contribute through advisers, continuous Slack-based red teaming, and close research translators; ChatGPT for Healthcare underwent nine testing waves over six months. HealthBench spans 5,000 conversations and about 49,000 criteria, while a randomized study at Kenya’s Penda Health produced statistically significant improvements in diagnosis and treatment outcomes.
The bottleneck is shifting from medical knowledge toward context acquisition, multimodality, and product integration. Singhal says text performance is already strong outside some subspecialties, but models work best when given the complete record and still need better access to imaging, voice, longitudinal measurements, and subtle physical signals. The likely architecture combines native multimodal representations with tool use, Python, and specialized models: “the value of data that [people] collect on themselves should increase over time as model intelligence increases.”
OpenAI expects clinical adoption to accelerate through 2026, with entrenched workflows—not physician protectionism—the main constraint. ChatGPT for Healthcare adds HIPAA compliance, medical-evidence retrieval, and clinician workflows; it launched with eight leading institutions and generated more inbound demand than the team could handle. Singhal’s goal is for AI-assisted care to become “part of the norm of care” by year-end, though he cautions that healthcare changes over months, not weeks.
Privacy is being treated as adoption infrastructure: health data receives added encryption, remains segregated from ordinary ChatGPT activity, and is not used to train foundation models. Singhal does not claim more data lacks value; he argues that lowering users’ “activation energy” through a clear privacy bargain will create greater long-term health impact. The next policy frontier may be consented data sharing for trial matching, N-of-1 treatments, and Labenz’s proposed “AI and the right to try.”
Healthcare also serves as OpenAI’s applied alignment laboratory, but Singhal does not claim the hardest oversight problem is solved. Medical models already outperform individual physicians in narrow areas, forcing work on scalable oversight, calibrated uncertainty, expert aggregation, and model character; meanwhile, chain-of-thought has not shown broad drift into “neuralese” as reinforcement learning scales. The upside case is medical “Move 37s”—unexpected diagnoses or treatments that raise the ceiling of health—but the balance between rapidly expanding capability and rarer safety failures remains uncertain.