DeepSeek V4 Flash

DeepSeek V4 Flash is a Mixture-of-Experts (MoE) language model with 284B total parameters and 13B activated per token, released as a preview alongside the larger DeepSeek V4 Pro. It targets highly efficient long-context intelligence, supporting a context window of up to one million tokens. Flash reaches reasoning quality comparable to the Pro version when given a larger thinking budget, making it a strong general-purpose model for reasoning, coding, math, and agentic workflows.

Architecture

DeepSeek V4 Flash pairs a sparsely activated MoE design with a hybrid attention mechanism that combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to improve long-context efficiency. Manifold-Constrained Hyper-Connections (mHC) strengthen residual connections and stabilize signal propagation, and the model was trained on more than 32T tokens using the Muon optimizer for faster convergence. Weights use an FP4 + FP8 mixed format, with MoE expert parameters stored in FP4 and most remaining parameters in FP8.

The model exposes three reasoning modes: Non-think for routine daily tasks and low-risk decisions, Think High for complex problem-solving and planning, and Think Max for pushing the boundary of model reasoning.

Key Features

• Mixture-of-Experts model with 284B total parameters and 13B activated per token
• Context window of up to one million tokens for long-document analysis and retrieval
• Hybrid CSA and HCA attention for efficient long-context serving
• Three selectable reasoning modes spanning fast responses to deep deliberation
• Native tool-calling and reasoning parsing for agentic applications
• Open-source under the MIT License

Use Cases

• Conversational assistants and chat applications
• Coding assistance, code generation, and software engineering agents
• Mathematical problem-solving and step-by-step reasoning
• Long-context document analysis, summarization, and retrieval over very large inputs
• Agentic workflows involving tool use, browsing, and multi-step planning

Benchmarks

In Max mode, DeepSeek V4 Flash reports strong reasoning and knowledge results, including 88.1 on GPQA Diamond (Pass@1), 86.2 on MMLU-Pro, and 34.8 on Humanity's Last Exam (Pass@1). It shows leading coding and math performance, with 91.6 on LiveCodeBench, a Codeforces rating of 3052, and 94.8 on HMMT 2026 Feb.

Long-context evaluations at the full 1M-token window report 78.7 on MRCR 1M and 60.5 on CorpusQA 1M. On agentic tasks, Flash reports 79.0 resolved on SWE-bench Verified, 56.9 on Terminal Bench 2.0, 73.2 on BrowseComp, and 45.1 on HLE with tools.

The base model, DeepSeek V4 Flash Base, reports 88.7 on MMLU (5-shot), 68.3 on MMLU-Pro, 90.8 on GSM8K, and 69.5 on HumanEval (Pass@1), among other results.

For full model details and the complete evaluation tables, see the model card on Hugging Face.