> much cheaper to access

I'm calling BS. I use Kimi, chatGPT (Astra/Sol), and Claude (Opus 5.5), and Kimi is by far the most expensive in terms of tokens and $/work done. Using OpenCode harness.