Deepseek models are good, no doubt on that... but what I have personally observed, as opposed to the benchmarks showing DSV4Flash-0731 ranking only below the kimi-k3 and near glm-5.2, and matching top models, is that that's not true.

At best, I personally would rate DSV4Flash-0731 being equivalent to gpt-5.6-luna-medium.

If they increase prices to anything more than 1.5x, I will stop using it. It's not worth.