Anthropic Drops Opus 4.8 + Robinhood Tests AI That Trades For You | AI News in 5

So users are already pushing back on Google's new AI search update. KeepSeek just slashed AI model prices by up to 75% and Robinhood is now testing AI agents that can trade stops in your behalf. This is your AI news 5 by the AI report family in Los Angeles. Get into it. Last week we covered Google's massive AI search update. Now we're seeing the first real user reaction to it. DuckDuckGo says installs of his browser and search tools increase sharply in the days following Google's announcement. The announcement in question, AI powered search features. According to TechCrunch, US app installs for the private browser was an average of 18% week over week between May 20th and May 25th, peaking at more than 30% on May 25th itself. DuckDuckGo also said traffic to his AI research page, which disables AI generated answers and images by default, claimed more than 22% week over week during the same time period. This story gives us one of the first measurable signs that some users are actively looking for alternatives to AI heavy search experiences rather than fully embracing them. I want to know where you land on this one. Do you prefer traditional search results? Or do you like the direction Google's taking? Let me know in the comments down below. Next up, DeepSeek just caught prices on its V4 Pro AI models by up to 75%. The company's latest pricing changes significantly reduce the cost of using its models, escalating what is becoming a very aggressive pricing battle across the AI industry. The cuts apply across both input and output token pricing making DeepSeek one of the cheapest frontier level AI providers currently available. This matters because low model costs directly reduce the barrier for companies building AI products and services. For start-ups and enterprise teams, model pricing has become one of the biggest factors in deciding which AI providers to build around. For now, the pressure is really on companies like OpenAI and Thropic and Google who are not just competing on performance but on cost as well. We'll see what happens next year. On with the third story, Apple just announced a new set of AI-powered features across the iPhone, iPad, Mac and Vision Pro and they're not aimed at chat or assistance but at accessibility and device control. On Vision Pro, one feature under development could allow certain users to control compatible power wheelchairs using eye tracking. On iPhone and iPad, Apple is adding features that can describe images, read detailed information from photos and documents and respond to questions about what the camera is seeing in real time. There's also updates coming to Apple's magnifier app which will be able to describe surroundings using Apple intelligence and respond to simple voice commands like zoom in or turn on flashlight. What ties all of this together is that Apple is applying AI less as a conversational assistant and more as an interface layer that helps people interact with their devices and environment in practical ways. Next up, Robinhood is now testing AI agents that can help users trade stocks through connected brokerage accounts. According to reports, the feature allows AI agents to place trades on behalf of users using dedicated wallets with preloaded balances rather than giving the systems unrestricted access to full investment accounts. Users will also receive notifications for trades made by the AI agent and will be able to monitor activity directly inside the Robinhood. At alongside, built-in fraud detection and oversight controls. This is part of a broader push towards AI agents that do more than just answer questions. Robinhood says the feature is currently being tested with limited access users as the company continues refining the system. Even with those safeguards in place, it still raises interest in questions around how comfortable people are becoming with AI systems making financial decisions on their behalf. I'm Jürgen Puyuthan on this one. Would you trust an AI agent to help manage or execute trades inside your investment account? Is this not what's being done and hedge funds at a large scale? All with our final story. Anthropic just released Claude Opus 4.8, the newest version of its flagship API model and it's available now to everyone across any planet. Here's everything you need to know about the update. One of the biggest improvements is that the model is more honest about what it does and does not know. Early testers found Opus 4.8 more reliable during complex agent-based tasks and less likely to confidently push forward with weak assumptions or unsupported conclusions. Alongside the model launch, Anthropic introduced several new features as well. Claude users on all plans and all models can now control how much effort gets put into a task, allowing the system to prioritise either fast responses or deeper reasoning depending on the situation. Claude is also getting a new dynamic workflows feature designed to help the system break down and manage much larger multi-step problems. We can expect more updates from Claude soon as the announcement included that Mythos class models may become more widely available in the next few weeks. I want to know what you think about Opus 4.8. Have you tried it? Have you noticed the change in honest air behaviour from the new model? Whatever it's said, do you find what's best for you? Low, medium, high, extra max. This was your A&U from 5 by the AI report. My name is Liam Lawson. I'll see you next week. If this hit for you, subscribe and share it with one person who needs to hear this. New episodes every week, see you on the next one.