Meta AI starts writing daily briefings, and Claude Opus 5 arrives at half of Fable 5's price

Meta's chatbot now writes a morning briefing from a person's calendar and keeps a weekly meal plan going without being asked twice. Meta says the assistant, running on its Muse Spark 1.1 model, can also connect to email and calendar apps, build slides, and carry a task through from start to finish. The rollout starts in a limited set of markets through the Meta AI app and meta.ai, with WhatsApp among the platforms Meta says will follow, which puts a working assistant in front of more than a billion people who open those apps every day.

Google moved the same idea down a price step on the same day. Spark, its assistant that keeps working after the laptop is closed, is now open to US subscribers on the $20 Google AI Pro plan; at launch it needed Ultra, which runs $100 to $200 a month. Spark reads Gmail and Calendar in the background, drafts replies, sorts what needs attention first, and asks before it takes a large action.

Both work from a standing instruction instead of a prompt, and that requires continuing access to the mail and calendar they draw on. Spark needs a personal Google account rather than a Workspace business account, so an employee who tries it does so on personal mail, outside anything an IT department administers. Meta lets a person set how far the assistant may act on their behalf, and offers an incognito mode for conversations a person wants kept private.

Google's own documentation notes that Spark can carry out bulk actions in Google Tasks without asking for confirmation.

Anthropic released Claude Opus 5 on July 24 at $5 per million input tokens and $25 per million output tokens, the same rates it charged for Opus 4.8. The company says the new model comes close to the intelligence of Fable 5, its most expensive one, at half the price, and it went live on all platforms the day it was announced. On coding and knowledge-work evaluations, among them Frontier-Bench and GDPval-AA, Anthropic reports Opus 5 as its strongest model; on cybersecurity tasks it stays behind Mythos 5.

A setting Anthropic calls effort controls how many tokens the model spends on a request. The developer documentation gives high as the default, xhigh for the most demanding work, and medium or low for routine tasks, so a team can run the same model cheaply on the ordinary half of its queue and at full strength on the rest.

Harvey, the legal AI company and an early customer, measured what that saves in practice. Its head of applied research, Niko Grupen, said Opus 5 matched what Opus 4.8 produced in its maximum-reasoning mode while generating 26% fewer tokens on average.

Most HR teams use AI to write. A new survey of 264 HR professionals found the majority use it for drafting and brainstorming, the tasks closest at hand, and go no further.

About a third have handed it something harder, a defined sequence of steps with fixed rules that AI runs from one end to the other while people handle the exceptions. A companion study of 1.4 million real workplace conversations with AI suggests why so few get that far. The skill that marks out the second group is managerial rather than technical: they frame the problem, steer the reasoning while it happens, and check what comes back.

On the survey's own measure, the teams that made that move report results two to three times better than the peers still drafting.

The processors inside most Android phones get more expensive on September 1. Qualcomm told customers in a letter that prices rise by a double-digit percentage on everything shipped after that date. Bloomberg reviewed the letter, sent on July 24, and Reuters reported it the same day without independently confirming it. Taiwan's Storm Media, which also obtained the letter, said it covers the product range as a whole, naming no individual product and no single rate.

Those chips also go into Windows laptops sold on their AI features, Meta's smart glasses, smartwatches and wireless earbuds. Qualcomm says it can no longer absorb the higher prices its own suppliers charge, and that it could not find alternative sources for the components.

In the copy Storm Media obtained, the reason given for the increase is supply capacity tightening under demand from AI and data centers.

The AI inside PowerPoint is no longer OpenAI's. Microsoft has released two models of its own, one that generates and edits images and one for voice, and switched them into its own products: PowerPoint, OneDrive, its contact center software, and its medical transcription tool. Microsoft says running its own models cuts what those features cost it.

For a company whose customer calls go through Dynamics 365, the voice a caller hears now comes from Microsoft's own model, under the same subscription as before.