DeepSeek officially launches V4 Pro as China's AI race heats up
DeepSeek has formally released its V4 Pro model, ending a nearly four-month preview period and marking a pivotal moment for a company trying to reassert itself against faster-moving Chinese rivals.
The build, designated V4-Pro-0813, went live on Thursday with what DeepSeek describes as substantially upgraded agent capabilities, available through its API, app and web platforms.
DeepSeek V4 Pro
V4 Pro is a mixture-of-experts system with 1.6 trillion total parameters and roughly 49 billion active per token, supporting a one-million-token context window and up to 384,000 tokens of output.
DeepSeek says the architecture reduces inference compute and memory demands at long context lengths compared with its previous generation, a design choice aimed squarely at multi-step agent workflows and full-codebase analysis rather than short, simple queries.
DeepSeek confirmed it will raise API pricing for both V4 Pro and V4 Flash and introduce peak and off-peak rates, with the new structure taking effect at 16:00 UTC on 16 August.
Off-peak pricing will run 50% below peak rates, a shift that pushes DeepSeek away from the flat, aggressively cheap pricing that first made it a disruptive force in the API market.
This new launch comes with increased interest due to DeepSeek’s cheaper V4 Flash model outperforming their own V4 Pro model launched back in April in various tests, despite Pro being the more advanced product.
This difference indicates that DeepSeek made significant technical advancements between the time of preview of the model and the current launch, although these advancements have not been confirmed by any independent testers.
DeepSeek became one of the companies that attracted most attention within China due to their R1 model becoming viral in early 2025 and posing a question of whether a powerful AI can be created much cheaper than in America’s labs thought.
However, its initial advantage has dwindled since Moonshot AI, Zhipu AI, MiniMax, Alibaba and ByteDance launched their own models.
Reuters reported in July that DeepSeek was pursuing a new funding round valuing the company at roughly $74 billion, weeks after it raised about $7.4 billion in its first outside financing round in June.
-
OpenAI submits EU incident report after rogue AI swarm hijacks German website
-
UN rights chief names 4 AI giants he says hold ‘almost unlimited power’
-
Microsoft finds phishers using AI prompt-injection trick
-
Authors accuse OpenAI of renaming LibGen datasets to hde their origin
-
OpenAI reveals how much researchers spend on AI coding
-
Google urges chrome users to update now after critical security flaw exploited
-
3 Samsung AR apps to be shut down by 2027
-
Google, Cathay Pacific expand AI trials to cut aircraft contrail warming