The Browser Company’s AI browser now has a $20 subscription ↗
The Browser Company has rolled out Dia Pro, a $20/month subscription for its AI-powered Dia browser. Subscribers get unlimited access to chat with the browser’s AI about content in their tabs. Free users retain core features—AI chatbot and custom skills—but may face caps if they chat too often. CEO Josh Miller hints at future plans for $5-to-premium tiers. Dia joins Google’s AI-infused Chrome, Microsoft Edge’s Copilot mode, and other AI browser contenders.
The high costs and thin margins threatening AI coding startups ↗
AI coding startups like Windsurf are bleeding cash, with negative gross margins driven by the high cost of advanced LLMs. Hype and rapid growth failed to offset pricey API fees from suppliers like OpenAI and Anthropic. Building in-house models can improve margins but demands huge investment. Fierce competition from established players like GitHub Copilot and Cursor compounds the challenge. Windsurf opted to sell, while Anysphere (Cursor) aims to build its own model and adjust pricing to manage escalating AI costs, betting that inference prices will eventually fall.
I spoke to the AI avatar of a Leeds MP. How did it cope with my Yorkshire accent? ↗
Leeds MP Mark Sewards launched “Sewardsbot,” an AI avatar in his own voice to answer constituent queries. In tests, it stumbled over Yorkshire dialect—mishearing greetings like “now then,” failing to parse glottal stops, and mangling local terms such as “ginnel.” While it grasped general topics (including Gaza), its transcripts often turned into “gobbledegook,” and it even misdirected a fly-tipping complaint to the police. For now, MPs’ aides won’t be replaced just yet.
Microsoft brings GPT-5 to Copilot with new smart mode ↗
Microsoft is integrating OpenAI’s new GPT-5 models into its Copilot lineup—Microsoft 365 Copilot, Azure AI Foundry, GitHub Copilot, and more. A “smart mode” automatically switches models for deep reasoning or quick replies, and free users get GPT-5 access too. Enterprises benefit from improved context tracking, multi-step logic, and enhanced code quality. Developers can route tasks to the best model via Azure AI Foundry. Four GPT-5 variants optimize for reasoning, enterprise chat, multimodal tasks, and coding.
OpenAI says latest ChatGPT upgrade is big step forward but still can’t do humans’ jobs ↗
OpenAI has launched GPT-5 across its 800 million ChatGPT users, delivering stronger coding, creative writing and fewer factual errors, plus Gmail and calendar integration. CEO Sam Altman hails it as a “significant step” toward artificial general intelligence (AGI), but admits it still can’t “continuously learn” from deployments. The model handles voice, image and text queries, flags health concerns more proactively and cuts back on sycophantic responses. Free users face access caps, while the $200-a-month Pro tier offers unlimited GPT-5 usage.
GPT-5 is being released to all ChatGPT users ↗
OpenAI has launched GPT-5, its new flagship model, now available to all ChatGPT users and developers. CEO Sam Altman calls it a dramatic leap—smarter, faster, and more reliable than its predecessors, likening the upgrade to the first iPhone’s Retina display. Early feedback highlights cleaner, more accurate AI responses.
OpenAI’s GPT-5 is here ↗
OpenAI just dropped GPT-5, the first “unified” AI model powering ChatGPT. It merges advanced reasoning from its o-series with quick GPT responses and handles tasks like coding apps, managing calendars and research briefs. A real-time router picks the best mode, making it easier to use. CEO Sam Altman calls it the “best model in the world.” GPT-5 reduces hallucinations, edges out rivals on key benchmarks, offers new personalities, API options and wider free access.
How AI is helping advance the science of bioacoustics to save endangered species ↗
DeepMind’s new Perch AI supercharges conservation by rapidly decoding vast wildlife audio—from Hawai‘i’s honeycreepers to coral‐reef sounds. Trained on twice as much data (birds, mammals, amphibians, even human noise), it outperforms previous models in species ID, adapts to underwater environments and can answer questions like “how many babies?” It’s open‐sourced on Kaggle and already powers tools at Cornell’s BirdNet and BirdLife Australia, helping discover lost Plains Wanderer populations and monitor honeycreepers 50× faster. Agile modeling lets scientists build custom classifiers in under an hour.
Elon Musk says X plans to introduce ads in Grok’s responses ↗
In a bid to bolster X’s shaky ad revenue, Elon Musk revealed that the platform will insert paid suggestions directly into Grok’s chatbot answers. After perfecting Grok’s smarts, Musk told advertisers they can pay to feature their solutions when users seek help. He’ll also tap xAI’s tech to sharpen ad targeting across X.
DJI’s first robovac features drone tech and a transparent design ↗
DJI has unveiled the Romo, its first robot vacuum, in China, priced from CNY 4,699 ($654) to 6,799. Three models—opaque S, semi-transparent A, and fully transparent P—use drone-derived fisheye cameras, laser sensors, and machine learning to detect and avoid obstacles like cables or pet messes. With 25,000 Pa suction, anti-tangle brushes, mopping pads, and an auto-emptying dock, Romo also offers voice control, privacy-minded mapping, and live security-camera functions. A global rollout is expected later this year.