Google on Tuesday announced Health AI updates at its annual The Check Up event, aimed at healthcare use cases, including improved overviews in Google Search for health queries, medical records APIs in Health Connect, and new health-focused "open" AI models to improve the efficiency of AI-powered drug discovery.
Also Read: Oracle UK Investment, ServiceNow AI Agents, Google AI Chip, Tech Mahindra–Google Cloud Partnership
1. Google Health AI Updates
In Search, Google said it is using AI and "best-in-class quality and ranking systems" to expand "knowledge panel" answers on thousands of health-related topics, and adding support for healthcare queries in Spanish, Portuguese, and Japanese, starting on mobile. Search already provided knowledge panel answers for common health topics such as the flu or the common cold, but the update greatly expands the number of topics the knowledge panels cover, Google said.
The company is also adding a separate feature in search labelled "What People Suggest," which aims to provide users with information from people with similar lived medical experiences.
"Using AI, we're able to organise different perspectives from online discussions into easy-to-understand themes, helping you quickly grasp what people are saying. For example, a person dealing with arthritis might want to know how others with this condition exercise," said Karen DeSalvo, Chief Health Officer at Google, in a blog post on March 18.
With this feature, Google says users can quickly uncover real insights from people who also have the condition, with links to click out and learn more. "What People Suggest" is available on mobile devices in the US.
Medical Records APIs
Google on Tuesday also launched new medical records application programming interfaces (APIs) globally for its Health Connect platform for Android devices. The APIs enable apps to read and write medical record information like allergies, medications, immunisations and lab results in standard FHIR format. With these additions, Health Connect supports over 50 data types across activity, sleep, nutrition, vitals and now medical records to enable users to connect everyday health data with data from their doctor's office, the company said.
AI Co-Scientist
The company recently launched what it calls an AI co-scientist, a new system built on Gemini 2.0 that aims to act as a "virtual scientific collaborator" to researchers and scientists.
Google said the AI co-scientist helps researchers parse large amounts of scientific literature to generate new hypotheses. "For instance, let's say researchers want to better understand the spread of a disease-causing microbe. They can specify this research goal using natural language, and the AI co-scientist will propose testable hypotheses, including a summary of relevant published literature and a possible experimental approach," DeSalvo explained.
The AI co-scientist, according to Google, is designed to help experts uncover new ideas and accelerate their work. The company said it is already working with partners, including Imperial College London, Houston Methodist, and Stanford University, to see how researchers use this tool and plans to launch a trusted tester program.
TxGemma
Google also announced TxGemma, a collection of Gemma-based open models that the company hopes will help improve the efficiency of AI-powered drug discovery. Google said TxGemma is capable of understanding regular text and the structures of different therapeutic entities, like small molecules, chemicals and proteins.
"This means researchers can ask TxGemma questions to help predict important properties of potential new therapies, like how safe or effective they might be," DeSalvo said. TxGemma is set to be released in the coming weeks.
Capricorn AI Tool
At its event, the company also detailed its work on an AI tool called Capricorn, developed in collaboration with the Princess Maxima Center for Pediatric Oncology in the Netherlands.
"It uses Gemini models to help physicians accelerate the identification of personalized cancer treatments by combining vast public medical data and de-identified patient data."
"Based on its analysis, Capricorn rapidly generates summaries of treatment options and relevant medical publication, which allows physicians to have more in-depth discussions on how to achieve the best possible health outcomes for their pediatric patients. With AI, physicians have more time to dedicate to what’s most important: patient care," Google explained.
Advancing healthcare with AI
Earlier this month, Google said AI is improving health outcomes globally, noting that it has developed AI models to help detect diseases, including breast cancer, lung cancer and diabetic retinopathy.
"Over the next decade, our health-tech partners in India and Thailand aim to deliver 6 million diabetic retinopathy screenings at no cost to patients, and Apollo Radiology International will build on our AI models to provide 3 million free screenings across India for tuberculosis, lung cancer and breast cancer," Yossi Matias, Vice President and Head of Google Research, said in a blog post on March 4.
In May 2024, Google announced Med-Gemini, a family of Gemini models fine-tuned for multimodal medical domain applications. For mobile and wearable devices, Google announced in June 2024 the Personal Health Large Language Model, another fine-tuned version of Gemini, designed to interpret sensor data and generate insights and recommendations about an individual’s sleep and fitness patterns.
Also Read: AI: Google Gemma 3, Palantir to Work With Archer, Qualcomm, and Anthropic-CommBank Partnership
2. xAI Acquires GenAI Video Startup
Elon Musk's AI company, xAI, has acquired Hotshot, a startup working on AI-powered video generation tools to compete with OpenAI's Sora. Aakash Sastry, Hotshot's CEO and co-founder, announced the news in a post on X on Monday.
"Over the past 2 years we’ve built 3 video foundation models as a small team — Hotshot-XL, Hotshot Act One, and Hotshot," Sastry wrote. "Training these models has given us a look into how global education, entertainment, communication, and productivity are about to change in the coming years. We're excited to continue scaling these efforts on the largest cluster in the world, Colossus, as a part of xAI!"
Hotshot said on its site that it began sunsetting new video creation on March 14. Existing customers will have until March 30 to download videos they've created using the platform, the company added.
xAI Grok 3
On February 19, xAI unveiled the latest version of its chatbot, Grok 3, which Elon Musk called "the smartest AI on Earth." The company then announced the release of two beta reasoning models, Grok 3 (Think) and Grok 3 Mini (Think).
"Trained on our Colossus supercluster with 10x the compute of previous state-of-the-art models, Grok 3 displays significant improvements in reasoning, mathematics, coding, world knowledge, and instruction-following tasks," xAI said in a blog post announcing Grok 3.
Also Read: AI: Celestial AI Funding, WNS Acquires Kipi.ai, CoreWeave OpenAI Deal, Perfios Buys CreditNirvana
3. Mistral Releases New Small AI Model
French artificial-intelligence (AI) startup Mistral AI unveiled a new open-source model on Monday, March 17, that the company says outperforms comparable models like Gemma 3 from Google and GPT-4o Mini from OpenAI, setting the stage for increased competition in a market dominated by US tech companies.
The model, called Mistral Small 3.1, processes both text and images with 24 billion parameters—a fraction of the size of leading proprietary models—while matching or exceeding their performance, according to the company.
"Mistral Small 3.1 is the first open-source model that not only meets, but in fact surpasses, the performance of leading small proprietary models across all these dimensions," the AI company said.
"Building on Mistral Small 3, this new model comes with improved text performance, multimodal understanding, and an expanded context window of up to 128k tokens," Mistral announced in a blog post. The firm claims the model processes information at speeds of 150 tokens per second, making it suitable for applications requiring rapid response times.
Mistral said Small 3.1 can run on a single RTX 4090 or a Mac with 32GB RAM, making it a great fit for on-device use cases. The model can be fine-tuned for specialised domains, creating accurate subject matter experts. This is particularly useful in fields like legal advice, medical diagnostics, and technical support.
The new model is designed for a wide range of enterprise and consumer applications that require multimodal understanding. Potential use cases include document verification, diagnostics, on-device image processing, visual inspections for quality control, object detection in security systems, image-based customer support, and general-purpose assistance.
Also Read: Mistral AI Unveils New Models for On-Device AI Computing
Mistral OCR
Earlier in March, Mistral announced Mistral OCR, which the company describes as the "World's best document understanding API."
Mistral OCR is an Optical Character Recognition (OCR) API capable of extracting text, tables, equations, and images from complex documents. Mistral claims that technology will transform how organisations process and utilise vast repositories of information.
According to the company, Mistral OCR processes up to 2000 pages per minute, supports multilingual and multimodal capabilities, and delivers structured outputs such as JSON for integration into AI workflows.
According to internal tests, Mistral OCR leads the market in text extraction accuracy, particularly for scanned documents, mathematical content, and multilingual text. Unlike traditional OCR solutions, it also extracts embedded images, making it ideal for scientific research, regulatory filings, and historical document digitisation.
Mistral said OCR is already helping enterprises and research institutions digitise literature, streamline customer service, and preserve historical archives. Additionally, OCR has also been helping companies convert technical literature, engineering drawings, lecture notes, presentations, regulatory filings and much more into indexed, answer-ready formats.
Mistral OCR capabilities are free to try on le Chat and the company expects the model to improve further in the coming weeks.
