Qwen 27B maxes out at a 16GB context. A nice thing about DeepSeek V4, especially Flash, is that its context size stays tiny even at 1M tokens! Which in turn opens up wide batching on common consumer platforms.