DeepSeek V4 hits public availability — reasoning and agentic upgrades land in open weights, with API pricing undercutting Qwen3-Max
DeepSeek's V4 model — first previewed in April with claims of reasoning parity against OpenAI, Anthropic, and Google frontier models — is now in public open-weights availability. The Hangzhou-based lab pitches major upgrades in reasoning and agentic capability that act autonomously on user behalf, distributed under the OSS license that made V3 the breakout China story of 2025.
The substantive piece is the price-quality curve. Qwen3-Max trades at $1.20 input / $6 output per million tokens on Alibaba Cloud and leads Arena-Hard at 90.5; DeepSeek V4 enters the leaderboard at a similar score with a more aggressive price ceiling. For the open-weights buyer that runs inference on owned hardware, V4 lands as the highest-capability OSS option since Llama 5.
The strategic frame is the China-OSS-frontier convergence. DeepSeek V4 + Qwen's continued 1M-context expansion + Kimi's agent-framework leadership put three China labs at the frontier of distinct capability axes (reasoning, multilingual, agents). For US enterprises subject to EU AI Act August 2 obligations, OSS provenance is the procurement-friction path — V4's release timing is favorable.
Semafor — China's DeepSeek launches new AI model → · CNN Business — China's AI upstart DeepSeek drops new model. Will it make waves like last year? → · RadarAI — Top China-Built AI Models to Watch in 2026 →