AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know

The past week has been a whirlwind of AI developments, with major announcements from industry giants like OpenAI, Anthropic, and Google. From new AI assistants and powerful language models to innovative editing tools and a White House summit, the landscape of artificial intelligence continues to evolve at a rapid pace.

OpenAI's DevDay Highlights: DOT, GPT-6.1 Soul, and More

OpenAI's DevDay was packed with significant reveals, the most prominent being DOT, their new AI assistant agent. Described as "remarkably capable" and "always-on," DOT is designed to handle a wide range of tasks without requiring users to select specific models. It proactively monitors communications like emails and Slack messages, offering to draft responses or flag urgent items. Early access users noted its impressive proactiveness, with DOT automatically identifying tasks and providing personalized summaries of daily agendas by sifting through various communications. However, DOT comes with a steep price tag, accessible only to Pro and Business Premium users, with plans starting at $100 per month, making it a tough sell compared to free alternatives like Meta's Muse, despite DOT's superior underlying models and extensive integrations with tools like Notion, To-Doist, Slack, and Google services.

Another major announcement was GPT-6.1 Soul, a new, more economical model that OpenAI positions as nearly as good as Astra. Priced at $2 per million input tokens and $10 per million output tokens via API (compared to Astra's $10 input, $50 output), Soul offers comparable performance in coding tasks and is a more cost-effective option for many applications. It's available to paid plan users (Pro, Business, Enterprise, Edu) but not yet in chat mode, only in work mode and Codecs. Benchmarks show Soul scoring just one point below Astra on Artificial Analysis, making it a highly economical choice at around 72 cents per task.

OpenAI also teased an Ultra Fast Mode, leveraging Cerebrus chips to generate 300 tokens per second—eight times faster than standard mode. This speed comes at a premium, priced at 6x the cost and exclusively available on the $500 per month plan. Other notable announcements included Private Intelligence for confidential computing, Codecs in the Cloud for persistent coding tasks, the Decisions API (similar to Jev for finite decision-making), Agents API supporting computer use, easier Plugin Extensions for developers, and new collaborative features like ChatGPT Spaces and Pages for team collaboration and document creation within ChatGPT. They also introduced a Meetings Plugin (a competitor to Granola), Sign in with ChatGPT for developers to offload API costs to users, and a new Pro 500 tier with access to Ultra Fast Mode and an OpenAI Marketplace.