健康是人们使用ChatGPT最有意义的方式之一。每周，超过2.3亿人向ChatGPT咨询健康与保健问题：理解健康信息、解读化验结果、为就诊做准备、处理保险事宜、培养更健康的生活习惯，以及弄清楚下一步该问什么。
借助GPT-5.5 Instant，我们在健康领域取得了显著进步，包括在识别何时可能需要紧急护理、询问相关背景信息、解释不确定性以及让复杂信息更易于理解方面都有改进。在我们最具挑战性的健康评估中，GPT-5.5 Instant的表现现已达到与我们前沿思考模型相当的水平。由于所有ChatGPT免费用户均可使用该模型，更多人能够从这些改进中受益。
这一进步既反映了模型能力的提升，也体现了我们健康评估背后由医生主导的工作。在我们的各项努力中，一个由全球医生组成的网络通过审查模型回复示例、描述理想行为以及识别失败模式，帮助定义在真实健康场景中何为"良好"。与医生合作使我们能够衡量健康领域的进展，并持续改进ChatGPT的回复方式。
衡量健康领域的进展
在健康领域，进步意味着提供准确、易懂且基于良好判断的回复：识别何时需要更多背景信息，解释不确定性而不夸大信心，并帮助人们了解何时应寻求医疗护理。
为了衡量这一进展，我们使用了专门的健康评估，包括HealthBench和HealthBench Professional(在新窗口中打开)。这些评估使用真实的健康对话和医生编写的评分标准，来评估准确性、安全性、沟通能力、背景意识、完整性和适当的升级处理等品质。
GPT-5.5 Instant在包括HealthBench Professional在内的健康评估综合指标上，达到了与我们最新前沿模型相似的健康表现，相比GPT-5.3 Instant有了显著提升。5.5 Instant（2026年5月发布）和5.3 Instant（2026年3月发布）可供ChatGPT所有免费用户使用（受限于使用限制），我们使用API定价来计算5.4 Thinking和5.5 Thinking的成本。
作为另一项对比，我们请医生为具有代表性的健康对话撰写回复，他们拥有无限时间和互联网访问权限（但不使用AI）。随后，一个独立的医生小组将这些医生回复与模型回复随时间进行对比，审查在真实互动中重要的品质，包括准确性、沟通能力、完整性、指令遵循程度以及健康决策帮助性，共审查了3500条回复。
在此评估中，GPT-5.5 Instant的回复在所有标准上均被评为优于医生撰写和旧版模型的回复。
医生们认为，GPT-5.5 Instant回复的失败模式少于旧版模型和医生。例如，与旧版模型和医生相比，GPT-5.5 Instant在未根据当地医疗背景进行调整、遗漏危险信号或转诊建议、或未在必要时向用户寻求更多背景信息等情况更少。
鉴于我们的模型在健康领域的使用规模，衡量近期模型改进的另一种方法是监测生产流量。我们使用保护隐私的监控器来追踪健康回复中可能的事实性问题。根据对近期健康领域生产流量（每周数十亿条消息）的比较，在过去两个月中，至少有一个被标记的事实性问题的回复率下降了71%。
更好的回复是什么样的
比较模型在真实健康问题上随时间变化的回复，可以看出ChatGPT在健康相关方面取得了哪些改进：识别何时情况可能需要紧急关注，以更好的判断处理不确定性，以及为人们提供更清晰、更有用的下一步行动指导。
进步背后的医学专业知识
这一进步是由医生们塑造的，他们帮助我们定义、衡量和改进ChatGPT中的健康回复。
OpenAI与一个由260多名医生组成的全球网络合作，这些医生来自60个国家，使用49种语言，涵盖26个医学专业。他们的反馈指导着ChatGPT如何回应各种场景下的健康问题，从日常保健问题到更复杂的临床情况。
医生们审查模型回复示例，并评估其是否准确、清晰、完整、适当谨慎且有用。他们帮助识别回复中可能遗漏重要背景信息的地方，可能听起来过于自信的地方，应该更明确说明下一步行动的地方，或者应该更直接鼓励某人寻求医疗护理的地方。
迄今为止，医生们已经审查了超过70万个模型回复示例，这些示例反映了患者和临床医生在现实世界中如何使用ChatGPT。每隔几分钟，就有一位医生审查一条新的回复。他们的反馈成为评分标准和评估准则，帮助研究人员衡量回复在真实健康场景中是否准确、安全、清晰、完整、适当谨慎且有用。这为我们提供了一种更清晰的方式，来观察模型在哪些方面变得更好，哪些方面仍需改进。
将健康改进带给更多人
这项工作也支持了OpenAI在健康领域更广泛的工作，包括为医疗保健构建的工具，例如面向临床医生的ChatGPT和面向医疗保健的OpenAI，这些工具通过文档处理、研究和护理交付等任务为医疗专业人员提供支持。
改善人类健康将是AGI最个人化、最切实的影响之一。随着我们的模型不断改进，我们的目标是让ChatGPT在这些时刻更加准确、更有用、更具影响力——并持续将这一进步带给更多人。
Health is one of the most meaningful ways people use ChatGPT. Every week, more than 230 million people turn to ChatGPT for help with health and wellness questions: making sense of health information, understanding lab results, preparing for appointments, navigating insurance, building healthier habits, and figuring out what to ask next.
With GPT‑5.5 Instant, we’re seeing a substantial step forward in health, with improvements in recognizing when urgent care may be needed, asking for relevant context, explaining uncertainty, and making complex information easier to understand. On our most challenging health evaluations, GPT‑5.5 Instant now performs at a level comparable to our frontier Thinking models. Because it is available to all free users in ChatGPT, more people can benefit from these improvements.
That progress reflects both advances in model capabilities and the physician-led work behind our health evaluations. Across our efforts, a global network of physicians helps define what “good” looks like in real-world health situations by reviewing example model responses, describing ideal behavior, and identifying failure modes. Working with physicians gives us a way to measure progress in health and improve how ChatGPT responds over time.
Measuring progress in health
In health, progress means delivering responses that are accurate, understandable, and grounded in good judgment: recognizing when more context is needed, explaining uncertainty without overstating confidence, and helping people understand when to seek care.
To measure that progress, we use health-specific evaluations, including HealthBench and HealthBench Professional(opens in a new window). These evaluations use realistic health conversations and physician-written rubrics to assess qualities like accuracy, safety, communication, context awareness, completeness, and appropriate escalation.
GPT‑5.5 Instant reaches health performance similar to our latest frontier models on an aggregate of health evaluations, including HealthBench Professional, substantially improving from GPT‑5.3 Instant. 5.5 Instant (released May 2026) and 5.3 Instant (released March 2026) are available for all free users in ChatGPT (subject to limits), and we use API pricing to calculate cost for 5.4 Thinking and 5.5 Thinking.
As another comparison, we asked physicians to write responses for representative health conversations, with unlimited time and access to the internet (but not AI). A separate panel of physicians then compared these physician responses with model responses over time, reviewing qualities that matter in real interactions, including accuracy, communication, completeness, instruction following, and health decision helpfulness, across 3,500 reviewed responses.
GPT‑5.5 Instant responses were rated higher than physician-written and older model responses across criteria in this evaluation.
Physicians rated GPT‑5.5 Instant responses as having fewer failure modes than those from older models and physicians. For example, GPT‑5.5 Instant had fewer instances of not tailoring to local healthcare context, missing red flags or referral to care, or failing to seek additional context from the user when needed than both older models and physicians.
Given the scale of usage of our models in health, another way to understand recent model improvements is to measure production traffic. We use privacy-preserving monitors on production traffic to track possible factuality issues in health responses. Based on a comparison of recent production traffic in health—billions of messages a week—the rate of responses with at least one flagged factuality issue has fallen by 71% in the last two months.
What better responses look like
Comparing responses from models on real-world health questions over time shows how ChatGPT has improved in ways that matter for health: recognizing when a situation may need urgent attention, handling uncertainty with better judgment, and giving people clearer, more useful guidance about what to do next.
The medical expertise behind the progress
This progress is shaped by physicians who help us define, measure, and improve health responses in ChatGPT.
OpenAI works with a global network of more than 260 physicians across 60 countries, 49 languages, and 26 medical specialties. Their feedback informs how ChatGPT responds to health questions across a wide range of scenarios, from everyday wellness questions to more complex clinical situations.
Physicians review example model responses and assess whether they are accurate, clear, complete, appropriately cautious, and useful. They help identify where a response may miss important context, where it may sound too confident, where it should be clearer about next steps, or more directly encourage someone to seek medical care.
To date, physicians have reviewed more than 700,000 example model responses that reflect how patients and clinicians use ChatGPT in the real world. Every few minutes, a physician reviews a new response. Their feedback becomes rubrics and evaluation criteria that help researchers measure whether responses are accurate, safe, clear, complete, appropriately cautious, and useful in real-world health situations. This gives us a clearer way to see where models are getting better and where they still need work.
Bringing health improvements to more people
This work also supports OpenAI’s broader work in health, including tools built for healthcare, such as ChatGPT for Clinicians and OpenAI for Healthcare, which support medical professionals with tasks like documentation, research, and care delivery.
Improving human health will be one of the most personal, tangible impacts of AGI. As our models continue to improve, our goal is to make ChatGPT more accurate, more useful, and more impactful in those moments — and to keep bringing that progress to more people.
本文内容采集自官方网站，排版和翻译可能与原页面存在差异。
阅读官方全文