OpenAI launches ChatGPT Health as Google tests AI Inbox for Gmail

OpenAI introduced ChatGPT Health on January 7, 2026, as a dedicated space in ChatGPT for health questions. Users can connect medical records and wellness apps so responses take their personal health information into account.¹ Access begins with a small group through a waitlist, with broader availability on the web and iOS planned in the coming weeks.¹²

OpenAI keeps Health conversations, files, and memories separate from regular chats. The company says content from this space is not used to train its foundation models.¹³ ChatGPT Health is meant to support medical care, not provide diagnoses or replace treatment.¹ According to a figure published by OpenAI, more than 40 million people ask ChatGPT health questions daily; that is not a usage figure for the new product.⁴ Available connections also depend on the app and region.¹⁵

Google announced AI Inbox for Gmail on January 8, 2026.¹ The new view brings together “Suggested to-dos” and “Topics to catch up on,” highlighting tasks and important email updates to help users prioritize.¹² Access is initially limited to select trusted testers in the US, with broader availability planned for the coming months.¹

Google is also expanding AI search in Gmail and tools for writing and revising emails.¹ The search feature is designed to answer questions about a user’s inbox directly.¹ Availability matters for businesses evaluating these tools: the new capabilities are starting in English in the US, and some advanced features require a paid subscription.¹³ AI Inbox itself is not yet a generally available default view.¹²

NVIDIA introduced Alpamayo at CES 2026 in Las Vegas on January 5 as a family of open AI models, simulation tools, and datasets for autonomous vehicle development.¹ Its Alpamayo 1 model takes video input and produces driving trajectories along with reasoning traces for its decisions.¹² The model is available on Hugging Face, and the release also includes the AlpaSim simulation framework.¹²

The approach is intended to help developers address rare and unusual driving situations that are difficult for autonomous vehicles.¹ The reasoning traces give development teams another way to examine planned maneuvers.¹² NVIDIA presented Alpamayo as a development resource, not a general launch of a feature in production vehicles.¹ Its announcement does not set out uniform commercial-use terms for every component.¹³

On January 6, 2026, xAI announced that it had closed an upsized $20 billion Series E funding round.¹ The company plans to use the money to expand its computing infrastructure, accelerate AI product development and deployment, and support research.¹ That work includes Grok 5, which Reuters reported was still in training.² xAI also cited the expansion of large GPU clusters as a goal.¹

The round illustrates the capital required to build computing capacity for large AI models. xAI did not disclose a valuation or specify how much of the funding was equity versus debt.¹ It also gave no release date or new benchmark results for Grok 5.²¹ The funding gives xAI more resources for infrastructure and development, but it does not establish how its next model will perform.

Amazon has launched Alexa.com as a browser interface for Alexa+.¹ Customers in Alexa+ Early Access can sign in with their Amazon account and chat with the assistant.¹ That gives them a way to use Alexa+ without speaking to it. Conversations can also continue across devices.¹ Amazon has not announced general access for all Alexa users.¹

The website extends Alexa+ alongside voice and the mobile app; it is not a separate assistant.¹² A browser chat interface puts Alexa+ closer to services such as ChatGPT and Gemini.³⁴ Amazon also emphasizes tasks beyond answering questions, including shopping, smart home controls, and planning.¹ Early Access users can now test how well those tasks work through the browser.

Stanford Medicine introduced SleepFM on January 6, 2026, as an AI model that estimates long-term disease risk from one night of clinical sleep recordings.¹ The accompanying Nature Medicine study evaluated more than 1,000 disease categories and found reasonable predictive performance for 130. Outcomes assessed include dementia, heart attack, and stroke.¹² Its input is sleep-lab polysomnography, not data from consumer wearables.¹²

SleepFM is designed as a foundation model that combines signals from a sleep study rather than only identifying sleep stages.² The findings come from retrospective clinical records, however, and do not establish that deploying the model improves treatment decisions.²³ The study does not demonstrate prospective validation in routine care or regulatory clearance for clinical use.¹²

Alibaba introduced Qwen3-VL-Embedding and Qwen3-VL-Reranker for multimodal retrieval on January 7, 2026.¹ The open-source models are designed to handle text, images, screenshots, and video: the embedding model maps content into a shared vector space, while the reranker scores the relevance of retrieved results.¹² The models are available through Hugging Face, GitHub, and ModelScope; deployment through the Alibaba Cloud API was announced as forthcoming.¹

Built on Qwen3-VL, the models are intended for uses including visual search and multimodal RAG.¹² They provide components for search applications, not a finished consumer search product.¹ Companies considering them should test retrieval quality on their own documents and media: published performance claims do not establish how well the models will work on company-specific data.¹

Z.ai, formerly known as Zhipu AI, began trading on the Hong Kong Stock Exchange on January 8, 2026.¹² The Chinese large language model developer raised about HK$4.35 billion in its IPO.² Shares were priced at HK$116.20 and opened at HK$120.¹²

Z.ai is regarded as the first Chinese company focused on large language models to list in Hong Kong.² The listing gives investors a way to buy shares in a Chinese model developer directly, rather than gaining exposure only through broader technology companies.² Its shares rose on the first trading day, although that gain does not establish how profitable the model business will become.² According to Bloomberg, the proceeds are intended largely for research and development.³

Anthropic is planning a funding round of about $10 billion at an approximately $350 billion valuation, according to media reports.¹² CNBC reported on January 7 that the company had signed a term sheet for the round.³ That does not mean the financing has closed, and the final amount could still change.³²

The potential round comes amid intense competition for capital among AI companies.⁴² At the reported valuation, Anthropic would be worth nearly twice as much as it was in its previous funding round.¹² The reports rely on unnamed sources, and Anthropic has not publicly confirmed the terms.¹⁴ Whether or when the round will close on those terms remains unclear.⁴²

Cursor introduced “Dynamic Context Discovery” for its coding agent on January 6, 2026.¹ Rather than keeping large amounts of information in the prompt, the agent retrieves context when it needs it.¹ Cursor says the change applies to all models.² The approach stores long tool outputs as files and loads MCP tools on demand.¹

In an internal A/B test, total agent token use fell 46.9% for runs that called an MCP tool.¹ Cursor says the result was statistically significant but varied substantially with the number of installed MCP servers.¹ The figure is therefore not an average across all Cursor tasks. The approach could matter for teams connecting multiple tools, but the test does not establish how much any particular setup will save.¹