Apple opened WWDC 2026 on June 8 with a comprehensive rebuild of Siri, powered by a custom 1.2-trillion-parameter Gemini model licensed from Google for approximately $1 billion per year (MacRumors, Let’s Data Science). The licensed model is approximately eight times larger than the largest cloud model Apple developed independently (TechTimes). Cloud queries route through Apple’s Private Cloud Compute infrastructure so that Google receives no user data, while on-device tasks continue to run on Apple’s own models (Let’s Data Science). The rebuilt Siri ships as a standalone app with a chatbot-style interface supporting text input, image and document attachments, on-screen awareness, and multi-step command execution across apps (Tom’s Guide). iOS 27 also introduces a Siri Extensions framework that lets users designate Claude, ChatGPT, or Gemini as their default assistant, and exposes an App Intents layer so developers can make their application actions callable by any of those providers (Business Standard, MacRumors). Developer betas are available following the keynote, with public betas targeted for July and a full release in fall 2026 (Tom’s Guide).
Anthropic will retire the claude-sonnet-4-20250514 and claude-opus-4-20250514 model identifiers from the Claude API on June 15 at 9 AM PT with no grace period; API calls using those strings will return errors immediately after the cutoff, with Anthropic recommending migration to Sonnet 4.6 and Opus 4.8 respectively (Anthropic, MindStudio, Tygart Media). On June 7, OpenAI detailed GPT-Rosalind’s expansion into controlled research settings: the life-sciences model scored 63.2% on LabWorkBench, a new evaluation measuring wet-lab protocol reasoning, compared to 55.8% for GPT-5.5, while consuming 5.3% fewer tokens (Winbuzzer, OpenAI). OpenAI simultaneously extended its Rosalind Biodefense program to vetted U.S. government and public-health organizations for outbreak modeling and medical countermeasure development (Let’s Data Science). As of June 6, Google’s Gemini 3.5 Pro, previewed at Google I/O in May with a promised 2-million-token context window and Deep Think reasoning mode, remained in limited preview with no confirmed general availability date (TechTimes).