V4-Flash
#44 in Frontier LLMsdeepseek · v4 · since 24. April 2026 · 13× · last seen Jun 30, 2026
31
Momentum
DeepSeek-V4-Flash is a Mixture-of-Experts language model released as a preview on April 24, 2026, with 284 billion total parameters and 13 billion activated parameters per token. It is the faster, cost-efficient variant of the V4 series alongside the larger V4-Pro, using a hybrid attention architecture (Compressed Sparse Attention + Heavily Compressed Attention) for efficient processing of up to 1 million tokens of context. The model is text-only (no multimodal input), released under the MIT license with open weights on Hugging Face, and offered via the DeepSeek API and chat.deepseek.com.
Momentum trend
04.04.03.07.
Features
| Key Benchmark (%) | SWE-bench Verified: 79.0%; LiveCodeBench: 91.6; GPQA Diamond: 88.1 |
| Context Window (Tokens) | 1,048,576 tokens (1M), max. 384K output |
| License | MIT License (open weights) |
| Multimodality | Text only (no image/audio input) |
| Platform | DeepSeek API (OpenAI- & Anthropic-compatible), chat.deepseek.com, Hugging Face (open weights) |
| Price per 1M Tokens | $0.14 input (cache miss) / $0.0028 input (cache hit) / $0.28 output |
| Release Date | April 24, 2026 (preview release) |