DeepSeek Releases Official V4-Flash Model with Enhanced Agent Capabilities and Further Cost Cuts, Skips V4-Pro

DeepSeek released the official version of DeepSeek-V4-Flash on July 31, introducing significantly enhanced autonomous agent capabilities and further reduced API costs versus the April 2026 preview, Caixin Global reported. The release landed slightly behind DeepSeek's earlier informal timeline and notably arrived without the companion V4-Pro flagship model that had been previewed alongside V4-Flash in April — with reported performance gains attributed to extensive post-training refinement rather than architectural changes. The release intensifies China's domestic low-cost frontier-AI race just two weeks after Moonshot AI's Kimi K3 open-weight release (July 16-27): Bloomberg noted the emergence of competitive, cheap Chinese models (Kimi K3, DeepSeek V4-Flash-0731) has sharpened global attention on the pace of Chinese AI catch-up, coinciding with a wave of strong earnings from Chinese AI chip designers (Cambricon revenue +100% YoY H1 2026, Moore Threads guiding to +149% H1 revenue growth) as Beijing's domestic-substitution push accelerates local AI chip adoption.