DeepSeek V4 Flash
🧠 AI Modeldeepseek
Efficient MoE model with 284B params, 13B activated, 1M context window.
DeepSeek V4 Flash is the latest iteration in DeepSeek's line of Mixture-of-Experts (MoE) models, designed to balance performance and efficiency. The model has a total of 284 billion parameters, but only 13 billion are activated per token, making it significantly faster and more cost-effective than dense models of similar total size. It supports a massive 1,048,576-token context window, enabling applications like long-document analysis and multi-turn conversations without truncation. The model is optimized for text-only inputs and outputs, and includes features like frequency penalty, logit bias, logprobs, and reasoning. According to benchmarks, DeepSeek V4 Flash achieves competitive ELO scores (e.g., 1278 in 3d, 1197 in asciiart, 1263 in codecategories) and ranks among the top models in various categories. It is available through OpenRouter with straightforward usage via the API, making it accessible for developers and researchers.
💡Highlights
- ├─284B MoE, only 13B activated per token
- ├─1M-token context window
- └─$0.10/M input, $0.20/M output
🎯For
- ├─Developers
- ├─Researchers
- └─AI Engineers