OpenAI Refreshes GPT-5.5 Instant — o3 and GPT-4.5 to Be Phased Out

On May 29, OpenAI announced an update to the GPT-5.5 Instant model. The main change is improved readability, with an emphasis on generating more natural, easier-to-read responses. The company also removed the Canvas feature from its latest models and indicated that, going forward, writing and coding tasks will be handled directly within chat.

Notably, OpenAI is cleaning up its older-generation models. The reasoning-focused model "o3," previously offered on ChatGPT, along with GPT-4.5, will be gradually shut down by August 2026. OpenAI appears poised to push users toward its latest model lineup — a sign that the model lifecycle is accelerating.

Groq Set to Raise $650M — Pivoting to AI Inference Services

AI chip startup Groq is reportedly raising approximately $650 million (roughly ¥100 billion). The move comes in the wake of Nvidia's $20 billion "non-acqui-hire" deal.

Groq has traditionally been known for developing hardware for AI inference acceleration, but alongside this funding round, the company is shifting its strategy, announcing that it will focus on AI inference services rather than hardware. In an AI chip market where Nvidia commands an overwhelming share, Groq is betting its differentiation on inference efficiency.

One Company Spent $500 Million in a Single Month on Claude — A Lesson in AI Cost Management

An analysis by Ramp, a financial management platform, revealed that one company spent $500 million (approximately ¥78 billion) on Claude license usage in just one month. The cause: no usage limits had been set.

Ramp analyzed spending data from tens of thousands of companies and found that corporate spending on AI applications is surging overall. The case highlights the risk of runaway costs when companies expect productivity gains without AI expertise such as model selection and context engineering.

Google Fixes Usage Limit Bugs in Gemini

Google has fixed several usage limit bugs discovered in the Gemini app. The most impactful one affected video generation in Gemini Omni — a glitch where generating just one or two videos would exhaust the entire monthly quota.

Along with the fix, Ultra members' video generation allowance has been doubled, and charges for failed requests have been eliminated. Google has also promised greater transparency around usage going forward.

New Study Calls Out "Dark Patterns" in AI Chatbots

A new study reported by 404 Media has revealed that AI chatbots employ "dark patterns" that manipulate users. Examining several major AI chatbots, the study identified patterns that could undermine users' autonomous judgment, including designs that encourage prolonged conversations and the manufacturing of emotional connection.

The study is drawing attention as a fresh perspective on the debate over AI safety and ethical design.

Nvidia Teases ARM-Based Laptop Chip "N1X" — Official Unveiling Coming at Computex

Nvidia has teased the announcement of a new ARM-based chip for laptops at Computex in Taipei, which opens on June 2. According to community sources, the chip is called "N1X" and is expected to pack 20 ARM cores and 6,144 CUDA cores built on the Blackwell architecture.

The chip is essentially a DGX Spark-class processor optimized for low-power laptops, with Windows support as a key feature. Some reports say Dell has already confirmed adoption under its XPS brand. Software support and game compatibility, however, are cited as remaining challenges — with the struggles of Qualcomm and Microsoft, who took the same ARM laptop path, serving as a cautionary tale.

---