Skip to content
IIAIT

AI Today · 6 November 2026

Claude adds a read-only personal finance beta, and Google makes Gemini Omni 1.2 Flash cheaper

Anthropic lets US subscribers connect bank accounts to Claude, Google releases a cheaper video model with longer clips, OpenAI publishes commitments for communities near its data centres, and Moonshot AI makes the full Kimi K3.1 Pro weights available.

  1. 01

    Claude gets a Personal Finance beta with read-only bank access for US subscribers

    On 5 November Anthropic began rolling out Personal Finance in Claude as a beta for Pro, Max and Ultra subscribers in the United States. Users connect bank and credit card accounts through Plaid, and Claude can then group spending by category, point out duplicate charges, unexpected fees and rising bills, and explain financial terms. According to TechCrunch and MacRumors, access is read-only, so Claude cannot move, manage or invest money; it can see balances, credit limits and up to two years of transactions. Anthropic says conversations in Personal Finance are not used to train its models and, according to MacRumors, are not added to Claude's shared memory.

    What it means for you

    The feature is not available in Europe for now. It shows how assistants are moving towards direct access to sensitive accounts, so before connecting any such service, check exactly what data it can read, how long it keeps it and how you can disconnect it.

  2. 02

    Google makes Gemini Omni 1.2 Flash generally available, with 15-second clips and a 30% lower price

    Google announced on 5 November that its video model Gemini Omni 1.2 Flash is generally available in the Gemini API and on Google Cloud. Clips can now be set from 3 to 15 seconds, up from a maximum of 10 seconds in the previous version, and the price for 720p video is 0.07 dollars per second, which Google says is 30% lower. Follow-up edits such as changing the background or lighting work without regenerating the whole clip. The model is also used in Flow, Google Vids and Google Ads, 15-second videos are rolling out in the Gemini app to Google AI Ultra subscribers, and every output carries an invisible SynthID watermark.

    What it means for you

    If you make short product or training videos, draft at low resolution and upscale only the final version to keep costs down. Generated video still needs a human review of faces, logos and text, and you should label it as AI-generated where your platform or client requires it.

  3. 03

    OpenAI publishes a Community Compact for its future US data centre sites

    On 5 November, the day after the US midterm elections in which many local measures on data centres were on the ballot, OpenAI published a National Community Compact and a policy blueprint for what it calls compute campuses. For future sites it owns and operates, OpenAI says it will pay the full cost of the electricity it uses, not ask public officials to sign non-disclosure agreements except for narrow security and trade-secret protections, and publish information on noise, water and energy use. The commitments are voluntary, and The Decoder notes that much of OpenAI's capacity is built and held by partners, which the compact does not clearly cover.

    What it means for you

    This does not change how you use OpenAI's products. It does show that the energy and water use of AI is becoming a public and political question, which may matter if your organisation reports on the environmental impact of the digital services it buys.

  4. 04

    Moonshot AI makes the full Kimi K3.1 Pro weights available on Hugging Face

    Moonshot AI has published the weights of Kimi K3.1 Pro, a mixture-of-experts model with 2.8 trillion parameters in total, of which 104 billion are active for each token, and a context window of about one million tokens. The model had been available through Moonshot's API since late October, and the company had promised the weights by 4 November. Moonshot says it removed cyber-related data from training for the open-weight version. The custom licence allows commercial use, but companies selling model access with more than 20 million dollars in annual revenue need a separate agreement, and very large products must display the model's name; purely internal use is exempt from both conditions.

    What it means for you

    Running a model of this size requires a cluster of high-end GPUs, so for most organisations it is relevant mainly through hosting providers. If you consider it for sensitive work, check the licence, where the hosting provider runs it and whether it meets your data protection requirements.