DeepSeek has launched its v4 API, introducing two new models: DeepSeek-V4-Flash and DeepSeek-V4-Pro, along with detailed API documentation on its website for developers. These models feature a mixture-of-experts design and are compatible with OpenAI ChatCompletions and Anthropic APIs, allowing for easy integration into existing systems. Additionally, model checkpoints for both variants are accessible on Hugging Face, enhancing their usability for developers.

DeepSeek: DeepSeek is a Hangzhou-based Chinese AI company that specializes in developing open-source large language models rivaling top proprietary systems in reasoning, coding, and agentic tasks. It emphasizes cost-effective, high-performance AI accessible via API and open weights. DeepSeek has published official API documentation for its V4 models, introducing Flash and Pro variants with developer specifications and integration details.
DeepSeek-V4-Pro: DeepSeek-V4-Pro serves as the flagship model of the DeepSeek-V4 series, focused on advanced reasoning, agent capabilities, math, STEM, and coding performance comparable to leading closed-source models. Released in preview as open-source with API availability. It is featured in the official V4 API documentation published by DeepSeek, providing model specs for developer use.
DeepSeek-V4-Flash: DeepSeek-V4-Flash is the lightweight variant in DeepSeek’s V4 model family, optimized for faster responses and efficiency on simpler tasks while supporting extended context lengths. It is designed for high-volume, latency-sensitive applications. The news highlights its inclusion in the newly released V4 API documentation alongside full interface details for developers.

Architecture: The models employ mixture-of-experts design with instruct and base versions.
Open Weights: Model checkpoints for both Pro and Flash variants are available on Hugging Face.
API Compatibility: V4 models support OpenAI ChatCompletions and Anthropic APIs for straightforward integration.