AI This Week

Anthropic unveiled Opus 4.6, calling it its most capable model for enterprise and knowledge work. The release follows new Claude Cowork plugins aimed at productivity, legal, sales, and marketing tasks, pushing the company deeper into territory long owned by software vendors. Investors have already punished that sector amid concerns AI platforms will replace some software-as-a-service tools; Salesforce is down 25% this year, SAP 18%, Intuit 32%, and Thomson Reuters 30%. Anthropic says Opus 4.6 tops major benchmarks, beating OpenAI’s GPT-5.2 and Google’s Gemini 3 Pro, and returns more “production-ready” first drafts for documents, spreadsheets, and presentations. It can run agent tasks across dozens of tools and analyze filings and market reports.

‍

AI voice startup ElevenLabs said it raised $500 million at an $11 billion valuation, tripling its $3.3 billion mark from a $180 million Series C in January 2025. Sequoia Capital led the round, joined by Andreessen Horowitz, Iconiq, Lightspeed, Evantic Capital, and Bond; cofounder Mati Staniszewski said Nvidia invested in September. Founded in 2022 and based in London, ElevenLabs moved beyond text-to-speech into speech-to-text, sound effects, dubbing, music and conversation. It sells voice and chat agents for enterprises and tools for brands and creators, counting Time, Nvidia, Meta, and Salesforce as users. The company said it ended 2025 above $330 million in annual recurring revenue and is building toward an IPO.

Fitbit co-founders James Park and Eric Friedman have launched Luffu, a new app built around AI that pulls together a family’s health data and keeps a running watch for changes. The company says the system can spot trends across records, wearables, notes, and uploads, then flag potential issues and surface follow-up questions for a caregiver. It also works like a chat assistant, answering health-related prompts in everyday language while drawing from the household’s stored information. Luffu supports kids, adults, aging parents, and even pets, aiming at the person who manages appointments, medications, and paperwork. Users can add details with voice, text, or photos, and the app connects with Apple Health and Fitbit. Access starts via a waitlist.

Leaks point to a new Anthropic model branded Claude Sonnet 5, with an internal date string set to February 3, 2026. Whether that marks a public release or an internal milestone remains unclear, but it lands during the week of Super Bowl LX on February 8, as AI labs push for consumer attention against ChatGPT and Google’s Gemini. Early hands-on testing describes a non-thinking Sonnet 5 variant that looks competitive on math with frontier models and, in some workflows, produces stronger code than Claude Opus 4.5. A standout demo involved structured visual output, where an ASCII world map prompt generated an unusually complete, detailed result, alongside UI and rendering-focused tasks. The tested build lists a 128k context window.

Elon Musk’s SpaceX is acquiring his AI start-up xAI, the company behind the Grok chatbot, in a move that consolidates AI, rockets, space-based internet, and media under one roof. The companies did not disclose terms, but a source familiar with the deal said it values xAI at $125bn and SpaceX at $1tn, a level that would make SpaceX the world’s most valuable private company. Musk framed the merger as an “innovation engine,” with a near-term focus on launching AI satellites and a longer-term plan for space-based data centers to meet AI’s energy and compute demands. The deal follows Tesla’s $2bn investment in xAI and arrives as SpaceX considers an IPO.

OpenAI said it will retire several models from ChatGPT next month, including GPT‑4o, the model some paid users favored for its warm conversation style. The company launched GPT‑4o in May 2024 and faced backlash in August after briefly removing access following the release of GPT‑5, then restoring it and promising advance notice before any permanent retirement. OpenAI now says only 0.1% of users choose GPT‑4o daily and most people use GPT‑5.2, citing recent upgrades to personality, customization, and creative ideation. GPT‑4.1, GPT‑4.1 mini, and o4-mini will also exit ChatGPT. The API stays unchanged.

Mark Zuckerberg has teased a major AI rollout slated for 2026, including “agentic” commerce tools aimed at handling shopping-related tasks for users. The plans point to AI systems that can take actions, not just answer questions, with a focus on workplace and consumer use cases tied to buying and selling. The preview suggests an expanded set of products and integrations designed to move AI from chat into daily transactions and business workflows. Few specifics were released, but the timing and scope signal a broad product push. The announcement adds fresh momentum to the competition among major tech firms racing to ship action-oriented AI agents.

Anthropic has released a new suite of interactive applications that bring its Claude AI model directly into Slack and other workplace tools. The launch integrates the AI heavily into daily business operations. Instead of switching tabs to access the model, employees can now utilize Claude’s capabilities within their existing communication platforms. The system examines context from chat threads to complete tasks and answer queries instantly. This release marks a significant expansion for the company as it targets enterprise users. The suite includes specific tools designed to reduce friction in professional environments. By placing the model inside third-party software, Anthropic aims to capture the workflow market.

The United Arab Emirates has launched a home-grown “sovereign” open AI model in a bid to challenge dominant US and Chinese systems and boost its technological standing. The new model, backed by state resources, is designed to be openly accessible yet tailored to regional needs, reflecting ambitions to diversify the UAE’s economy and reduce dependence on foreign tech giants. Researchers say its performance positions it alongside some of the best existing open AI models, potentially attracting global developers and enterprises. The move underscores growing competition in the AI landscape as nations seek both technological leadership and economic advantage.

Microsoft has revealed its new Maia 200 AI accelerator chip, a piece of hardware the company says is three times more powerful than rivals from Google and Amazon. The chip is an inference powerhouse. It is designed to help AI models apply their knowledge to real-world situations with greater speed and efficiency. Microsoft is already deploying the Maia 200 in its Azure cloud data centers to improve next-generation language models and power services like Copilot. For end users, this could mean faster responses from AI tools. Scientists and developers using Microsoft’s platforms may also see a performance boost for large-scale projects, from advanced weather modeling to biological research.