DeepSeek-V4-Pro
DeepSeek-V4-Pro is a production API model (serving the 0813 checkpoint since August 13, 2026) with a 1.6T total / 49B active MoE architecture, 1M-token context, 384K max output, and 500-request concurrency. It leads open-source benchmarks in agent capabilities, world knowledge, and reasoning, is MIT-licensed, and offers tiered peak/off-peak pricing starting August 16, 2026.