Chinese AI startup DeepSeek has quietly launched an updated version of its flagship model, DeepSeek-V4-Pro-0813. While the model struggles to match top-tier rivals on standard AI benchmarks and faces criticism over pricing, it excels in niche cybersecurity tasks by outperforming competitors in vulnerability detection.
- V4 Pro ranks mid-tier on AI benchmark indexes, trailing leading competitors.
- Model excels in cybersecurity, detecting more vulnerabilities despite lower precision.
- Pricing higher than median market rates, disappointing some users familiar with V4 Flash.
What happened
China-based AI startup DeepSeek has released an updated flagship model, DeepSeek-V4-Pro-0813, following an earlier preview release in April and a more affordable V4 Flash model in July. The update was quietly posted on the company’s website and briefly promoted before being removed. Benchmark tests show the model scoring 53 on the Artificial Analysis Intelligence Index, putting it on par with Zhipu AI’s recent GLM-5.2 but behind competitors like Terra, Moonshot AI’s Kimi K3, and OpenAI’s GPT-5.6 series.
The model encountered specific difficulties with sandboxed terminal environments and generating complex Excel financial models, leading to some user dissatisfaction and reports of the model halting prematurely during lengthy coding tasks. Despite these challenges, DeepSeek-V4-Pro-0813 has stood out in cybersecurity applications, where it outperformed rivals in identifying system vulnerabilities, as reported by Belgian cybersecurity firm Aikido Security.
Why it matters
DeepSeek’s release illustrates how emerging AI providers are navigating competitive pressures in the crowded Chinese tech market. While the flagship V4 Pro model’s overall AI capabilities lag behind some global leaders, its cybersecurity strengths carve out a potentially valuable niche that could differentiate it from peers. This is particularly significant as cybersecurity remains a critical priority for enterprises worldwide.
Pricing disparities have fueled mixed reactions. At $0.435 per million input tokens and $0.87 per million output tokens, the V4 Pro model is more expensive than many established alternatives, including its own V4 Flash sibling. These cost considerations are influencing adoption decisions, emphasizing the ongoing tension between performance, price, and specialized use cases in the AI landscape.
What to watch next
DeepSeek is preparing to expand its product portfolio with the upcoming launch of DeepSeek Harness, a software framework designed to enable large language models to operate as autonomous AI agents. This platform could accelerate innovation within the Chinese AI ecosystem by providing developers with new tools to harness model capabilities more effectively.
Market observers will also be monitoring how DeepSeek responds to the critiques on performance and pricing. Increased transparency, detailed technical disclosures, and competitive pricing adjustments may be necessary for DeepSeek to strengthen its position against leading global models from entities like OpenAI, Anthropic, and Moonshot AI.