Overview
Torch built software to bring a person's scattered health records into one place for AI to use. OpenAI announced its acquisition on January 12, 2026. The four-person team joined the company behind ChatGPT Health. The transaction joined a small specialist product with a service that already reached people asking health questions.[1]
Torch's useful insight was that better answers require usable patient context. Its acquisition supports that product judgment. The public record does not establish a repeatable standalone business: revenue, retention, pricing, and acquisition costs remain undisclosed.
Founding Story
Expa identifies Ilya Abyzov, Eugene Huang, James Hamlin, and Ryan Oman as Torch's founders. Its portfolio links the company to torchapp.com and marks it acquired by OpenAI.[2] Contemporary reporting traces the team's connection to Forward Health, where they had worked before Torch.[3]
The surviving company announcement describes a familiar problem: patients hold only part of their health history, and clinicians lack time to review growing volumes of information. Torch concentrated on assembling that context.[4]
In his January 12 announcement, Abyzov wrote: “The Torch team and I are joining OAI.” In a separate post about the responsibility, he wrote: “It's a daunting responsibility.” These statements document the transition and its stakes; they do not establish the company's original founding date or early decisions.[5][6]
Timeline
- Before January 2026: Torch developed its health-record product and worked with beta users, according to Abyzov's acquisition thread.[7]
- January 7, 2026: OpenAI introduced ChatGPT Health.[8]
- January 12, 2026: Torch and OpenAI announced the acquisition.[1]
- July 23, 2026: OpenAI began a broader U.S. rollout of Health in ChatGPT, with permission controls for using connected information across conversations.[9]
What They Built
Torch gathered lab results, medications, and visit recordings for use in an AI health application. The company called its approach a “medical memory for AI.” Its intended benefit was continuity: a new question could draw on earlier records rather than treating every interaction as an isolated exchange.[1][4]
That product promise contains two separate jobs. One is obtaining records from different sources. The other is making those records useful without losing their meaning. The acquisition materials explain the goal, but disclose no independent accuracy study, supported-source audit, or detailed architecture. It would be premature to infer those capabilities from the word “memory.”
Market Position
Target Customers
Torch addressed people whose health information sat across multiple providers and consumer services. Expa describes connecting thousands of sources, but supplies no adoption figures or verified count of active integrations. That statement describes the intended product reach rather than customer traction.[2]
Market Size
No credible standalone market estimate or Torch revenue figure was established. OpenAI's health audience demonstrates broad interest, but it cannot be counted as Torch customers or willingness to pay for Torch.
Competition
The most consequential comparison is the acquirer's own product. ChatGPT Health launched with medical-record connections through b.well and support for services including Apple Health. That existing infrastructure means Torch should not be credited with creating every connector or health feature later offered by OpenAI.[8]
Business Model
Expa lists Torch in its portfolio. The surviving sources do not establish Torch's financing amount, consumer price, revenue, margins, or customer acquisition costs.[2]
CNBC's $60 million headline and The Information's roughly $100 million headline describe different reported totals. The Information's indexed excerpt separates immediate consideration from retention equity. Neither company published confirmed terms.[7][10] These reports cannot establish investor returns or equal payouts to founders.
Post-Mortem
A specialist capability met existing distribution
Torch's outcome illustrates a structural advantage of joining a platform. A small team can concentrate on a difficult context problem while a larger company supplies an existing user relationship. In its January 12 acquisition announcement, Torch said it could pursue its vision “at a bigger scale than we could have ever achieved on our own.”[4] That supports a distribution explanation for the transaction; it does not prove OpenAI could not have built similar software internally.
Product continuity required a change in interaction design
The sequel also tests a simple assumption about health software: giving sensitive information its own destination does not guarantee people will use that destination. OpenAI reported that more than 70% of health conversations among people with access happened outside the original dedicated experience. Its July update allowed connected health context in other conversations with permission.[9]
That is OpenAI's later product evidence, not a measured Torch result. It shows why record aggregation alone does not settle the user experience. People must be able to bring relevant information into the conversation they are already having. The same update says connected information can be incomplete or out of date, preserving a need to check important details against original records.[9]
Key Lessons
- Torch chose a specific problem in health AI: bringing scattered records together before generating answers. Its acquisition gives that choice a concrete outcome.
- An acquisition price cannot substitute for evidence about product economics. Torch's public record leaves the standalone business unproven.
- The July sequel shows how distribution and interaction design remain separate tasks. Existing users still needed clearer control over when connected information entered a conversation.
Sources
- Fierce Healthcare: OpenAI acquires healthcare startup Torch
- Expa: company portfolio and Torch founders
- TechCrunch: OpenAI buys Torch
- Torch: joining OpenAI
- Ilya Abyzov: acquisition announcement
- Ilya Abyzov: privacy and responsibility
- Techmeme: contemporaneous reporting and founder thread
- OpenAI: introducing ChatGPT Health
- OpenAI: launching Health in ChatGPT
- The Information: acquisition report; indexed excerpt only