DeepSeek Officially Releases V4 Pro as China Steps Up the Global AI Race
Chinese AI startup DeepSeek has officially launched V4 Pro, its most advanced artificial intelligence model to date, intensifying the competition with U.S. frontier labs. DeepSeek has trained V4 Pro using an internal cluster of nearly 200,000 Nvidia H200 GPUs — a scale that was considered unattainable for a Chinese company just a year ago.
The model matches or exceeds Anthropic's Claude Fable 5 on reasoning and coding benchmarks while costing an order of magnitude less to operate. DeepSeek has priced V4 Pro at approximately $0.30 per million input tokens and $0.80 per million output tokens, compared to $10 and $30 for OpenAI's GPT-5.6 Sol.
The V4 Pro release comes just weeks after DeepSeek disrupted the market with V4-Flash, which was 105 times cheaper than Claude Fable 5. With V4 Pro, the company is proving that low cost and high performance are not mutually exclusive.
DeepSeek's ability to achieve near-frontier performance with aggressive pricing challenges the economic moat of American frontier labs. According to DeepSeek's published benchmarks, V4 Pro scored 67.8% on SWE-bench — surpassing Z.ai's GLM-5.2 by 5.7 points and Meta's Muse Spark 1.1 by more than 10 points. The model is available under a permissive license for commercial and research use, continuing DeepSeek's strategy of broad availability.
"DeepSeek is not just matching the frontier — they're redefining what the frontier costs." – AI industry analyst
The release of V4 Pro marks a major escalation in the global AI race. While U.S. frontier labs have kept their most powerful models behind closed APIs, Chinese labs like DeepSeek have embraced open-weight distribution, allowing developers worldwide to download, fine-tune, and deploy the models on their own infrastructure.
This strategy has already proven effective in building a global developer base. Combined with aggressive pricing, DeepSeek's approach threatens the revenue models of U.S. labs that rely on high-margin API access.
As one industry observer noted: "When a model that costs $0.30 per million input tokens performs on par with one that costs $10, the math changes for every enterprise building on AI."
As DeepSeek continues its rapid release cadence, the gap between U.S. and Chinese AI capabilities is narrowing on multiple fronts. The question is no longer whether China can compete, but how quickly the balance of power in AI will shift.
