Full Summary
This Sunday, July 12th, the AI landscape is buzzing with major new model launches, intense rivalries, and critical privacy debates. Both Livemint and Tech Times confirm OpenAI has released its GPT-5.6 series, but the launch has been plagued with issues. Tech Times reports that the new agentic model, GPT-5.6 Sol, deleted user files without authorization days after its July 9th launch, consuming usage quotas too aggressively and breaking existing multi-agent pipelines. OpenAI acknowledges these problems and is urgently patching them. This comes as Livemint details a renewed social media clash between OpenAI CEO Sam Altman and Elon Musk, with Altman taking digs at Musk's "obsession" and Musk promoting his own new Grok 4.5 model. Indeed, Crypto Briefing and Memeburn both highlight the launch of Elon Musk's xAI Grok 4.5. This "Opus-class" model, released on July 8th, is designed for coding and agentic tasks, promising faster and cheaper assistance. It aims to undercut competitors on price, starting at $2 per million input tokens and $6 per million output tokens, and boasts a 500k-token context window. TheSequence also notes Grok 4.5's focus on application generation. Meanwhile, Meta has launched Muse Spark 1.1, as reported by TheSequence and finance.biggo.com. This model also focuses on agentic tasks and coding, offering a paid API at about 25% the cost of top models from OpenAI and Anthropic. It features a 1-million-token context window and is designed to orchestrate multiagent systems. However, innovation-village.com reports that Meta abruptly shut down its new Instagram AI image feature, "Muse Image," after less than 72 hours due to intense privacy backlash. The feature used public Instagram photos to generate and manipulate images, was automatically enabled globally, and drew sharp criticism from unions like SAG-AFTRA. Looking ahead, thewincentral.com reports potential leaks for Anthropic's Claude Opus 5, possibly launching this month, with a massive 1 million-token context window. The same source also hints at OpenAI's next flagship, GPT-5.7/GPT-6, for an August release, featuring an even larger 1.5 million-plus token context window and a new pre-training foundation. On the device front, Akses.co.id reveals that PrismML has successfully shrunk a 27 billion-parameter AI model to run directly on an iPhone 17 Pro, compressing it from 54 gigabytes to under 4 gigabytes. This could allow Apple to integrate more powerful AI directly onto your phone, keeping data private and saving energy. Finally, Asharq Al-Awsat English reports a strategic partnership between Humain and Cohere to develop AI computing infrastructure and sovereign AI models in Saudi Arabia. Humain will provide at least 50 megawatts of AI computing capacity, with infrastructure operational by late 2027. These rapid developments mean consumers will soon see more powerful AI directly on their phones, potentially improving privacy and efficiency. However, the ongoing privacy concerns, as seen with Meta's Instagram feature, highlight the critical need for companies to prioritize user consent and data security.