DeepSeek V3 (0324)

DeepSeek-V3-0324 is the March 2025 checkpoint of DeepSeek-V3, released as a minor update that delivered substantial gains. It carries 685 billion total parameters with 37 billion active per token and supports a 163,840-token context window for long-document processing. The refresh reused the V3 base while improving the post-training pipeline, lifting benchmark scores notably on reasoning and coding, including MMLU-Pro (75.9 to 81.2), GPQA (59.1 to 68.4), AIME (39.6 to 59.4), and LiveCodeBench (39.2 to 49.2). At release it ranked among the strongest open-weight non-reasoning models on independent benchmarks. It remains a direct-answer model rather than a chain-of-thought reasoner, with strengthened tool-use support, making it well suited to coding assistants and agentic workflows that prioritize fast, capable responses without proprietary lock-in.