DeepSeek AI
DeepSeek-V4-Flash-Vision-Exp
DeepSeek-V4-Flash-Vision-Exp is an experimental multimodal model that extends the DeepSeek-V4-Flash architecture with visual modules for enhanced visual understanding. It maintains high-performance text-only agent capabilities while significantly improving multimodal agent performance across various benchmarks.
DeepSeek Harness
DeepSeek Harness (dsh) is an open-source agent runtime framework designed for building autonomous agents. It utilizes a modular architecture where every component—including models, tools, session management, and the agent loop—is implemented as a plugin powered by the Cordis framework.
DeepJIT
DeepJIT is a lightweight, header-only C++20 runtime library designed for Just-In-Time (JIT) compilation of kernels on NVIDIA CUDA GPUs and HUAWEI Ascend NPUs. It provides a unified interface for compiling, caching, and launching device kernels, enabling developers to manage JIT infrastructure efficiently within their C++ or Python extensions.
DeepSeek V4 Pro DSpark
DeepSeek V4 Pro DSpark is an advanced mixture-of-experts (MoE) language model optimized for high-efficiency, million-token context intelligence. It features a specialized speculative decoding module to accelerate inference performance while maintaining state-of-the-art reasoning and agentic capabilities.
DeepSeek-V4-Flash-0731
DeepSeek-V4-Flash-0731 is an official, re-post-trained version of the DeepSeek-V4-Flash model, featuring significantly enhanced agentic capabilities and a built-in DSpark speculative decoding module. It is designed for high-efficiency, large-scale tasks, offering competitive performance in coding, reasoning, and multi-step tool use.
clangd-ascend
clangd-ascend is a specialized version of the clangd language server tailored for AscendC kernel development. It provides essential IDE features such as code completion, diagnostics, and navigation for .asc files, supporting AscendC builtin types, address spaces, and kernel launch syntax.
DeepSeek-V4-Flash
DeepSeek-V4-Flash is a Mixture-of-Experts (MoE) language model developed by DeepSeek-AI, featuring 284 billion total parameters with 13 billion activated, and supporting a context length of one million tokens. It offers advanced capabilities in text generation and conversational AI, optimized for efficiency and scalability.
DeepSeek-V4-Pro
DeepSeek-V4-Pro is a Mixture-of-Experts (MoE) language model developed by DeepSeek AI, featuring 1.6 trillion total parameters with 49 billion activated parameters. It utilizes a hybrid attention architecture combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA), achieving 27% of single-token inference FLOPs compared to its predecessor, DeepSeek-V3.2, at a 1 million-token context length. The model supports a context window of 1 million tokens, enabling efficient processing of extensive text inputs. DeepSeek-V4-Pro is available under the MIT license, allowing for both commercial and non-commercial use. It is accessible through platforms like Hugging Face and NVIDIA NGC, with integration options for various libraries and inference providers. The model has been evaluated on multiple benchmarks, demonstrating strong performance across tasks such as reasoning, coding, and general language understanding. For instance, it achieved an 87.5% score on the MMLU-Pro benchmark and a 90.1% score on the GPQA Diamond benchmark. These evaluations highlight DeepSeek-V4-Pro's capabilities in complex reasoning and problem-solving tasks. The model is optimized for deployment on various hardware platforms, including NVIDIA GPUs and Huawei's Ascend AI processors, reflecting DeepSeek AI's commitment to supporting diverse hardware ecosystems. This versatility ensures that DeepSeek-V4-Pro can be effectively utilized across different computational environments, catering to a wide range of applications in natural language processing and AI-driven tasks.