Pradeep Viswanathan

ChatGPT already has a popular Memory feature that allows it to remember useful information about users, including their preferences, projects, and constraints, which will be used in context to provide more relevant responses. Now, ChatGPT on macOS is taking the memory concept to the next level with its new Computer History feature. In April, OpenAI's Codex team revealed a research preview feature called Chronicle that builds memories automatically from a user’s recent screen context.

Back in June, when OpenAI first announced the GPT-5.6 series, the company said that Sol would also launch on Cerebras in July, delivering speeds of up to 750 tokens per second for select customers. However, OpenAI did not reveal any details in July about running Sol inference on the Cerebras platform. On X, OpenAI employees later confirmed that the company had been privately previewing this high-speed mode with select customers.

Image: Google Back in 2024, when LLMs were still in their early stages, OpenAI, Google, and other model developers were engaged in a price war. Model prices kept falling for several months before eventually plateauing. In 2025, however, the trend reversed, with several companies increasing the prices of their latest models. Now, the trend is shifting once again, largely due to the growing popularity of Chinese AI models.