also its pretty big model inference costs are high even with margins running a 2.8T model costs a lot. if they release oss may be it goes down to $10-12 per million tokens.