DeepSeek-V4 is Here. So is Everybody Else.

DeepSeek released two models — V4-Pro (1.6T/49B active params) and V4-Flash (284B/13B active) — rather than a single flagship V4. The Pro tier's real-world API pricing (~$1.74/M input tokens) is significantly higher than V3's disruptive debut, making competitors like Qwen3.6, Kimi K2.6, and Xiaomi's MiMo-V2.5 Pro look equally attractive on cost. Meanwhile, frontier labs didn't stand still: OpenAI shipped GPT-5.5, Google released multimodal Gemma 4 with 256K context and local deployment capability, and Anthropic launched Claude Opus 4.7. V4-Pro's genuine strengths include a 1M context window with DeepSeek Sparse Attention and improved multi-step reasoning in Think mode. The overall takeaway: V4 is a solid model that arrived in a far more competitive market than previous DeepSeek releases, and 'genuinely good' is now just the baseline.