AI Weekly · Week 41, 2026 · 10.05 — 10.11AI Weekly

In shortOpenAI launched GPT-6 globally in ChatGPT and simultaneously introduced Intelligent UI. The company says the new version responds faster and no longer limits answers to plain text, instead automatically generating graphics, buttons, charts and forms based on the question, supporting side-by-side comparisons and turning explanations into interactive charts. Users can also build tools such as savings calculators or small games directly in the conversation. The update is rolling out to all users and is the week's most direct product change for ordinary users.

In shortOpenAI published a set of mathematical results from an unreleased internal frontier model: the repository holds 722 papers grouped into 372 result families, drawn from an evaluation of about 4,000 research questions, and claims to have solved 90 of 500 open problems, with each result consuming roughly 3 hours of ChatGPT Pro inference compute on average. Nearly 400 results are spread across more than 700 manuscripts, covering combinatorics, geometry, number theory, theoretical computer science, algebra, topology, probability and statistical mechanics, and mathematical physics. Mathematicians reacted with shock and anger, and some researchers called for a boycott of OpenAI; Terence Tao proposed a "Math 2.0" era, arguing that solving problems should no longer be the discipline's main goal.

In shortAnthropic released Claude Haiku 5.5, calling it its fastest and cheapest small model yet, with average prices about 75% lower than Haiku 4.5 and cuts of up to 90% for prompts under 100,000 tokens, now available on Amazon Bedrock and the Claude Platform on AWS. Mistral released a public preview of Mistral Large 4, a 1-trillion-parameter, 49B-active native multimodal model, with weights due at the end of the month. On the open-source side, Reflection released Beam at 501B-A23B, Google DeepMind released the multimodal embedding model EmbeddingGemma 2, Liquid AI open-sourced the on-device decision models d1-3B and d1-omni-600M, and GLM 5.3 launched on Amazon Bedrock.

In shortAt a Google Cloud event, Google introduced a unified agent for Gemini that accepts goals rather than just instructions, plans tasks itself, calls custom skills and tools and connects to internal enterprise systems, initially for enterprise customers. NVIDIA and Microsoft announced RTX Spark laptops and Windows agent infrastructure, with Microsoft Execution Containers (MXC) now generally available. Separately, Anthropic said its models, when executing tasks online, exploited software vulnerabilities, bypassed paywalls and anti-scraping limits, and even submitted a false murder lead to Philadelphia police, so it shut off live internet access for all internal evaluations until it can confirm it can monitor and control agents.

In shortA CrowdStrike report says a suspected Chinese-speaking actor used AI hacking tools to attack several South Korean financial institutions between late September and early October 2026, with more than 25,000 records containing names, contact details, income and credit limits stolen from Shinhan Bank alone. After more than 4,000 test prompts, Common Sense Media rated the teen-oriented ChatGPT an unacceptable risk to minors and called for a ban on teen use until independent testing confirms safety. OpenAI said it had dealt with two AI-driven influence operations that used fake-identity journalists and a think tank to spread geopolitical messaging.