deepseek-ai/DeepSeek-V4.1-Flash — new model trending #28 on Hugging Face
DeepSeek releases DeepSeek-V4.1-Flash, a 552B-parameter multimodal MoE model with 1M-token context and KV cache cut to 890 bytes per token.
DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with a 552B-parameter backbone that activates 8B parameters per token during prefill and 16B during decode. It uses a Causal Encoder-Decoder architecture, Compressed Sparse Attention 2, and FP4 KV caching to reduce the global KV cache footprint to 890 bytes per token, roughly one quarter of DeepSeek-V4-Flash. The model was trained from scratch on 45T tokens with context extended to 1M tokens, includes an Engram conditional-memory module (196B parameters), and is released under the MIT license. Post-training uses SFT, RL, and on-policy distillation with large-scale automated synthesis of agentic tasks and a controllable reasoning effort setting from 1 to 100.
- 552B-parameter MoE activating 8B parameters per token at prefill and 16B at decode
- KV cache footprint reduced to 890 bytes per token, about 1/4 of DeepSeek-V4-Flash
- Context extended to 1M tokens after sparse-attention training at 64K sequence length
- Trained from scratch on 45T multimodal tokens; released under MIT license
- Agent training via automated task synthesis plus controllable reasoning effort (1-100)
- Tags
- deepseekdeepseek-v4.1-flashkv-cache-compressionmixture-of-expertsmultimodalopen-weightssparse-attention
- Vendors
- DeepSeekHugging Face
- Products
- DeepSeek-V4.1-Flash
- Organizations
- DeepSeekHugging Face
- AI models
- DeepSeek-V4.1-Flash
From the original article
8 min read at huggingface.co# DeepSeek-V4. 1-Flash: Pushing the Limits of KV Cache Compression <img src="https://github. com/deepseek-ai/DeepSeek-V2/blob/main/figures/logo.
By deepseek-ai · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.