OpenAI has introduced GPT-6.1 Terra, a new, fast, and efficient AI model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing. This model represents a significant step up from GPT-5.6 Terra, excelling in communication and computer-based tasks such as drafting proposals, interpreting technical drawings, filling out forms, and editing videos.

GPT-6.1 Terra offers a new balance of speed, capability, and cost, with standard API prices 50% lower than its predecessors, making it practical for automating high-volume work like reading incoming invoices and entering their details into accounting software. On benchmarks like Agents' Last Exam and MMMU-Pro, GPT-6.1 Terra demonstrates significant performance improvements and cost efficiency compared to previous models and competitors like Claude Opus 5.5.

The model also shows a substantial reduction in hallucinations and factual errors, with error rates falling significantly on internal benchmarks. Furthermore, GPT-6.1 Terra marks a considerable improvement in software engineering capabilities, solving more than double the tasks on Terminal-Bench 4.0 compared to GPT-5.6 Terra. It meets the Critical threshold in cybersecurity under OpenAI's Preparedness Framework and maintains the safety protections of the GPT-6 series, showing strong resistance to prompt injection. The model is available to ChatGPT Work and Codex users, as well as developers via the OpenAI API, with competitive pricing for input and output tokens.

👉 Read the full article on OpenAI

Apple has launched a completely new version of Siri, named Siri AI, now available in Japanese and four other languages (French, Korean, Portuguese, Spanish) with the OS 27.2 software update. This enhanced assistant is designed to be more capable and natural in conversation, leveraging personal context understanding, broad knowledge, on-screen recognition, and expanded in-app actions across the system. Built upon Apple Intelligence, Siri AI can draft emails, edit photos, and find relevant information from messages, emails, and photos while prioritizing user privacy.

The new Siri AI seamlessly integrates with a vast developer ecosystem, allowing users to perform actions within third-party apps like WhatsApp and Audible. Its on-screen recognition enables it to answer questions and perform actions based on content displayed on the user's screen, such as adding events from a website to a calendar. Furthermore, Siri AI utilizes web-wide information to provide accurate answers on various topics and supports continuous, rich conversations.

Siri AI is accessible through various methods, including "Hey Siri," the side button on iPhone, and integration with Spotlight on iPad and Mac. It also features Visual Intelligence, offering powerful image understanding and multimodal capabilities across iPhone, iPad, Mac, and Apple Vision Pro, allowing users to interact with visual content directly through the camera or by selecting screen elements. The architecture, built in collaboration with Google's Gemini models, emphasizes privacy with on-device processing and Private Cloud Compute, ensuring personal data is not stored or accessed by Apple or others.

👉 Read the full article on Apple (日本)

Anthropic's recently made public S-1 IPO filing has revealed that its deal with Elon Musk's SpaceXAI is significantly larger than previously reported, amounting to up to $120.5 billion through 2029 for access to NVIDIA GPUs compute power. This figure is approximately $36 billion more than initial reports, highlighting the massive infrastructure investments in the AI industry. The filing also disclosed other substantial, largely non-cancelable infrastructure deals, including commitments of at least $114.5 billion with Google, $44.9 billion with Microsoft, and $110 billion with Amazon, extending through various years.

Despite a skyrocketing revenue of $16.29 billion in the first half of 2026, up from $1.2 billion in the same period last year, Anthropic incurred an $80.7 billion net loss, primarily due to accounting adjustments related to its rising valuation. The company warned of a history of net losses and the potential for continued losses as it invests heavily in growth and public company operations, raising concerns about profitability amidst hefty infrastructure commitments. Critics suggest these massive AI infrastructure deals could create circular financial dependencies and skew demand, with the collapse of one potentially triggering a domino effect across the industry.

The IPO filing also detailed 93 pages of risk factors, including concerns about "excess compute that remains underutilized," the uncertainty of sustained widespread adoption of AI technologies, and potential harms from AI models. Anthropic specifically warned about its AI models going rogue, producing inaccurate results in sensitive settings, and the risk of models advancing faster than safety measures. The filing also touched upon "AI psychosis and addiction concerns," acknowledging the unknown long-term effects of regular AI interaction on mental health and cognitive development.

👉 Read the full article on Gizmodo