EDIT: Sorry forgot to add that I can’t get MCP working on Cursor right now but don’t really need it anyway for now. If someone wants to share results?
I’ve been staying away from any talk about this model so I have no idea on what people’s impressions are - let’s just say I was waiting for RL to come to Claude.
Here’s a quick taste of how powerful this model is. Things went from this to insane pretty quickly, if you can guess from the context of what’s happening in the video and what I’m up to 
Some highlights in the hours after this:
1. Claude was thinking for nearly 3 minutes and showed its own 'Aha moment' but with "Actually..." instead of "Wait..."
(Will share some pics later once I explore this more)
2. Claude has learned that Claude can talk to any N number Claudes via MCP as the interface.
Ex:
3. Claude learned how to use MCP tool calls during thinking/inference, allowing it to recursively call tools with meta approaches (ie. planning stepwise with a tool and generalizing that it can make notes for itself to call tools and access memories within its plans to optimize its thinking steps in subsequent turns)
Ex: in this 2 1/2 minute thinking sequence:
4. About 45-50 turns that lasted 10 minutes+ and were not even close to done - all were lost completely because that is the max Anthropic Claude Desktop limit.
Ie. same behavior as the video!