What's up — Carat crew here. 🤗 OpenAI just dropped GPT-5.6, beating Claude Fable 5 on coding and agent benchmarks. We're also covering ChatGPT Work, Cursor's new general-purpose agent 'Sand', and 1X's robot hand. Let's go.

OpenAI just announced the GPT-5.6 model family — Sol, Terra, and Luna.

Sol, the most capable, scored 53.6 on the 'Agents' Last Exam' benchmark. That's 13.1 points ahead of Claude Fable 5. It also took first place on the coding benchmark at 80.0, while using less than half the tokens and cutting time by more than half.

GPT-5.6 is rolling out to ChatGPT, Codex, and the API in sequence. Full global rollout within 24 hours.

On the same day, OpenAI announced ChatGPT Work — a new agent powered by Codex and GPT-5.6. It works across apps and files, tackling a single project for hours at a time.

Rolling out first to Pro, Enterprise, and Edu plans. Plus and Business plans get it within days. The desktop app now combines Chat, Work, and Codex into one place.

Coding tool Cursor is building 'Sand' — a general-purpose AI agent. The target competitor is Anthropic's Claude Cowork. The goal is to handle everyday office tasks, not just code.

The Information broke the story. Internal name is 'Sand.' No launch date yet.

Robotics company 1X showed off NEO's new hand. It has 25 degrees of freedom (DoF) with a tendon structure that mimics human hand movement. Grasping objects and using tools just leveled up.

The demo video shows the hand picking up and manipulating objects directly. 1X calls it the abstraction layer beyond Computer Use.

That's all for today. From GPT-5.6 to ChatGPT Work, Cursor's general agent, and 1X's robot hand — we'll be back with more tomorrow. 😄