Augment Code: 22.5% of Our Users Are Consuming 20x What They're Currently Paying

I absolutely agree, Qwen Coder 480b is my go to, I'm also paying for OpenAI Plus/Codex and Cursor and none of those models ever give me an edge over Qwen. One can always setup an "ultrathink" model for when push-comes-to-shove, I use o3 through the api, but I seldom use it. Same applies to a large context (ie 1M) alternative, image processing etc. I'm doing an average of 120 M tokens/day that way and couldn't be happier.