Product Launch

Anthropic just made its most agentic mid-tier model the default for every free Claude user. The pitch is blunt, near-flagship autonomy at roughly a third of the price of its flagship.

OpenAI's strongest model yet is here, but you probably cannot use it. The US government asked OpenAI to gate GPT-5.6 to about 20 vetted partners first.

OpenAI just stopped renting all its compute and started building it. Meet Jalapeño, the company's first custom chip, designed from scratch with Broadcom for one job.

Adobe just turned its Semrush acquisition into a product. Brand Visibility measures whether AI engines name you and then deploys fixes, putting a real number on GEO.

Anthropic just put a virtual coworker inside Slack. Tag it in a channel, hand it a task, and it breaks the work into steps and finishes it on its own.

Databricks just walked into martech. CustomerLake puts a data platform where the CDP used to sit, and gives agents the context to run campaigns on their own.

Googling yourself is losing its meaning. A new site asks the real 2026 question instead, whether the major AI models can recall you at all without searching the web.

A government order took Fable 5 offline in an afternoon. Within days, a wave of downloadable coding models gave locked-out teams a second engine they cannot have recalled.

OpenAI just put a frontier security model and an open-source patching program in the same launch. The unspoken goal is to become too important for any government to switch off.

Almost every chatbot writes one word at a time. Google just open-sourced a model that drafts a whole paragraph at once, runs on a gaming GPU, and writes up to four times faster.