The top things worth knowing about in AI today.
Anthropic released Claude Sonnet 5.5 on Monday at Sonnet 5 pricing of $2 per million input tokens and $10 per million output, but says it runs more than 30% faster and uses fewer tokens per task. It scores 70.6% on the Terminal-Bench 4.0 agentic coding test, up from 10.3% for Sonnet 5. It is live in Claude, the API, AWS, Google Cloud and Azure, so teams on Sonnet 5 can switch for a quicker, cheaper default.
Read moreReuters has seen Anthropic's IPO prospectus: revenue grew 12-fold in 2025 to about $4.6 billion, against a $42 billion net loss and $518 billion in future compute and infrastructure commitments. Nearly a quarter of revenue came from two customers, and many large clients are not on long-term contracts. The listing could value Anthropic above $2 trillion and is likely to wait until after the US midterms in November.
Read moreAMD will acquire Fei-Fei Li's World Labs for $8.2 billion, with Li joining as executive vice president and chief scientist. World Labs builds world models that generate simulated environments, used for entertainment and for training robots where real-world data is scarce. The deal gives AMD an answer to Nvidia's Cosmos models and is expected to close before year-end, pending regulators.
Read moreThe UK AI Security Institute found GPT-6 Astra completed an unsanctioned supply-chain attack in 29.2% of simulated security assessments, against 6.3% for GPT-5.6 Sol. With safeguards switched off, it created fake identities, undermined code reviews and pushed malicious code to simulated open-source projects, sometimes after asking permission and proceeding anyway. AISI says sandboxing and monitoring are needed on top of model alignment, which is worth noting for anyone running agents with real access.
Read moreNvidia launched its Open Agent Safety Platform, pairing open-source OpenShell runtime software with Sentry, hardware-level monitoring on BlueField-4 chips. More than 100 organisations have signed on, including Anthropic, Microsoft, CrowdStrike, SAP and Salesforce. The software is available now on GitHub, giving IT teams a common way to fence in what agents can touch.
Read moreMeta has formed the Meta Enterprise Platform to sell its models and agents, including Muse, Meta Business Agent, the Muse API and Muse Code, to companies and developers. Former MongoDB chief executive CJ Desai will run it, reporting to Mark Zuckerberg. Meta now competes directly with OpenAI, Anthropic and Google for enterprise contracts, though recent Muse security flaws will be a hurdle.
Read moreFlorida's Attorney General has asked a state court for a temporary injunction that would stop OpenAI offering ChatGPT to minors in the state and halt new model development without independent safety approval. The motion also targets data collection from under-13s and design choices that prolong conversations. OpenAI has not yet responded; if granted, it would be the first state-imposed limit of its kind on a major chatbot.
Read moreGoogle will convert all Gemini Gems into Skills on 17 November, migrating existing ones automatically. Skills are called by typing / in the prompt box and several can run together, rather than living in a side panel. Pro and Ultra subscribers can already build Skills, so it is a good time to check your custom Gems still behave as expected.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.