The NXP LPC1110 starts at less than $1 for a 4KB flash part. That's cheaper than an AVR in my experience! I'm not really sure if it's all that more powerful but it's good enough for me. The thing I like is that it also has a single I/O line debugging, just like AVR's DebugWire, which isn't available on all parts and for which you need to purchase ATMEL's programmer since no one else has (AFAIK) implemented it.
Anyone have any experience with Cortex-M0 yet?