Haven't tried their API but I can tell you that their local model DeepSeek-V4-Flash-UD-IQ2_M is one of the best local models I've benchmarked. Despite being Q2 (!) it is still very very good and runs well on GB10 with 128GB VRAM. Another impressive recent model is Laguna S 1.2 .. about as good as DeepseekV4 in general but with more refined design sensibilties