Apple AI Updates: May 23, 2026

1. Apple Machine Learning Research drops VSAS-Bench for streaming vision-language models

Apple. Apple Machine Learning Research published VSAS-Bench on May 22, a benchmark for evaluating real-time visual streaming assistant models that continuously generate responses against an online stream of input frames rather than pre-recorded clips. The framework ships with over 18,000 temporally dense annotations across diverse input domains and introduces proactiveness and consistency metrics alongside synchronous and asynchronous evaluation protocols that vary memory buffer length and input resolution. A headline finding is that conventional VLMs adapted to streaming settings can beat purpose-built streaming models, with Qwen3-VL-4B surpassing Dispider by 3% under asynchronous evaluation and suggesting existing open models can transition to streaming workloads without retraining. Source

2. WWDC 2026 preview points to a Gemini-powered Siri chatbot redesign

Apple. A May 22 episode of The MacRumors Show, citing Bloomberg’s Mark Gurman, lays out a sweeping Siri redesign expected to land with iOS 27 at WWDC on June 8, repositioning Siri as a full ChatGPT-style chatbot with a dedicated app, persistent conversation history, and a “Search or Ask” entry point inside the Dynamic Island. The reporting says the new Siri will be powered by Google Gemini under the hood, an arrangement Apple is reluctant to emphasize publicly, with user-facing controls to auto-delete chats after 30 days, one year, or never. The episode also reiterates the Apple Intelligence-driven accessibility features previewed on May 19, including VoiceOver Image Explorer and natural-language Voice Control upgrades, framing them as a small taste of a much larger AI story at the keynote. Source