DeepSeek launches V4 Pro to challenge Anthropic's top AI model
What's the story
Chinese artificial intelligence (AI) firm DeepSeek has launched its latest model, the DeepSeek V4 Pro. The new version comes with improved agent capabilities and support for Responses Application Programming Interface (API) and Codex integration. The launch is part of a growing trend among Chinese and foreign AI companies to release more advanced models, making the competition even fiercer.
Multi-platform availability
How to access the new V4 Pro model
DeepSeek has confirmed that users can access the new V4 Pro model on mobile apps, web interface, and API by selecting "Expert Mode."
The company also updated its API documentation, replacing the preview version of its V4 Pro model with a new one, DeepSeek V4 Pro 0813.
This update marks the official release of the API version of DeepSeek V4 Pro.
Enhanced features
New model supports thinking and non-thinking modes
The updated model, DeepSeek V4 Pro 0813, comes with a 1 million-token context window and a maximum output length of 384,000 tokens. It also supports thinking and non-thinking modes.
The launch of this new model comes as AI developers are increasingly moving from simple chat functions to longer, multi-step workflows that require tool use, code execution, and persistent task completion.
Competitive edge
DeepSeek's benchmark results show competitive edge over global counterparts
DeepSeek has released benchmark charts showing strong performances across several agent-focused tests, including terminal operation, code engineering, tool use, and security-related tasks.
In some cases, the V4 Pro 0813 approached or even surpassed leading overseas models. The model's performance on benchmark tests is comparable to Anthropic's Claude Fable 5.
However, DeepSeek has priced its model competitively at 3 yuan ($0.44) per million input tokens and 6 yuan per million output tokens with cached input priced at 0.025 yuan.
Industry evolution
New frontier in AI model competition
The simultaneous releases of DeepSeek V4 Pro 0813 and Grok 4.6 indicate that the AI industry is entering a new phase of competition.
Frontier models are now defined not just by parameter scale or conversational ability but also by their ability to independently use tools, write and execute code, and complete multi-step tasks.
This shift highlights the growing importance of technical advances and attractive pricing for model providers in this evolving landscape.