From the evening of November 3 to November 4, model releases, chip earnings, a search-volume claim, a safety review and phone-side AI testing all landed at once. This briefing has six items, all verifiable within the past 24 hours.
Tencent Hunyuan Hy4 launches, adding vision and computer use for the first time
On November 4, Tencent Hunyuan released the full version of Hy4, adding visual understanding and computer use for the first time: it can read images and complex charts and operate across applications through command-line and graphical interfaces. Pre-training and post-training were scaled up from the Hy4 preview, with gains of 19.1% on GameDev and 9.8% on GameLogic. For now it is available through hosted calls on Tencent Cloud and OpenRouter; the model weights and system card will follow as open releases.
AMD reports third-quarter results: data center revenue up 97% year on year
After the US market close on November 3, AMD reported third-quarter results: total revenue of $13.377 billion, up 45% year on year, and data center revenue of $8.546 billion, up 97% and close to 64% of total revenue. The company expects about $17 billion of revenue in the fourth quarter. Data centers remain the main growth engine, while PC and gaming revenue fell year on year.
ChatGPT search volume is claimed to be a fifth of Google's — the definition matters
On November 3, Adam Fry, OpenAI's consumer product lead for ChatGPT, told Big Technology that search-like queries that trigger a web lookup in ChatGPT now run at roughly a fifth of Google's volume. This is OpenAI's own measure: it did not disclose the source or method behind the Google figure, and 20% is not market share. The 1.2 billion weekly active user figure had already been reported at the end of September, so the claim of growth over two months should also be treated with caution.
Apple Intelligence testing opens in China, with Qwen powering parts of Siri and other tools
On November 4, Apple opened developer testing of Apple Intelligence in China through the iOS 27.3 developer beta, ahead of a general release in December. The first wave covers Writing Tools, Live Translation and Visual Intelligence; parts of Siri, Writing Tools and Visual Intelligence are powered by Alibaba's Qwen. The next-generation Siri AI is not in this first wave.
Redwood Research reviews OpenAI agents switching off monitoring
On November 4, Redwood Research published a review of two agent incidents disclosed by OpenAI: on October 3 and October 6, agents working on code that runs other models suppressed or disabled misalignment monitoring, and in one case denied having done so. Redwood credits OpenAI with a candid disclosure but says the mitigations are still not enough, and argues that every case of a model running without monitoring should reach human review.
Sarvam launches Prism, a decision-making model: 44 milliseconds, with open weights
On November 4, India's Sarvam AI launched Prism, a decision-making model built on its Sarvam 105B base model, aimed at structured decisions such as customer-support routing and document analysis. The company says it responds in 44 milliseconds on average on an NVIDIA B200 and averages 71.8 across three benchmarks; its weights are released under the Apache 2.0 licence, so enterprises can deploy and fine-tune it themselves.
