Released
Feb 25, 2025
V2
Agent V2
V2 is the next step: more advanced agent behavior with stronger judgment about when to use a tool, plus the ability to understand images and richer inputs.
What’s improved from V1
- Tool awareness: recognizes when a tool is necessary vs when chat is enough.
- Better decomposition of tasks into steps and subgoals.
- More consistent formatting for deliverables and action plans.
- Multimodal understanding (e.g., interpreting screenshots or images in context).
Best use cases
- Research tasksWhere evidence and step-by-step reasoning matter.
- Tool workflowsLike “search → compare → recommend.”
- Image-based contextBugs, UIs, screenshots.
Positioning
V2 is where “agent” starts to feel real: not just chatting, but choosing actions. It’s the bridge between the foundation (V1) and the flagship “final boss” (01).
