DeepSeek V4 Pro: New AI Model Challenges Claude 3.5

China's DeepSeek has upgraded its V4 Pro artificial intelligence model. This new version is now available as an official API. What's interesting is that the V4 Pro shows major improvements in AI agent capabilities. Benchmark results indicate it's closing in on, and in some tests, even surpassing Claude Fable 5. The model now supports the Responses API and integrates with Codex. This expands its use for agent-based applications and coding tasks. DeepSeek has priced V4 Pro at 3 yuan per million input tokens and 6 yuan per million output tokens, with cached input costing just 0.025 yuan per million tokens. This aggressive pricing aims to lower the cost of advanced AI. The V4 Pro scored 87.9 on Terminal-Bench, just behind Fable5's 88. However, it surpassed Fable5 on CyberGym and AutomationBench. This launch highlights DeepSeek's effort to compete with global AI leaders, not just in intelligence, but also in AI systems' ability to use tools and complete complex tasks. This could mean more accessible and powerful AI tools for developers.