DeepSeek Launches V4.1-Flash to Make Large-Scale AI Inference Cheaper

Chinese AI startup DeepSeek has launched its latest V4.1-Flash model with 552 billion parameters and a context window of up to 1 million tokens, while activating only 8 billion parameters during input and 16 billion during output.

The company has made V4.1-Flash available through its API with native multimodal support. 

DeepSeek has retired V4-Flash and V4-Flash-Vision-Exp, while the older model names deepseek-v4-flash and deepseek-v4-flash-vision-exp will temporarily route requests to V4.1-Flash.

DeepSeek said multi-party testing found V4.1-Flash ahead of V4-Pro in performance, cost, speed, and total completion time, prompting the company to phase out the older model.

The company’s latest launch targets more efficient large-scale AI inference, using a Causal Encoder-Decoder architecture, sparse attention, and FP4 KV caching to reduce memory requirements. It brings the global KV cache footprint down to 890 bytes per token. The model consists of 20 encoder and 20 decoder layers, with a smaller KV cache.

The model supports both image and text inputs and lets users adjust its reasoning effort from one to 100. It also performs well on coding and agentic benchmarks.

According to DeepSeek, the model’s reduced KV cache can bring its memory requirements down significantly, with the company saying it uses one-quarter of the HBM and one-eighth of the SSD storage required by the previous generation.

An advanced MoE design with layers consisting of one shared expert and 384 routed experts, including six routed experts activated per token. The model has multimodal capabilities, and its DeepSeek-ViT vision encoder can convert images into visual embeddings, processing alongside text.

DeepSeek-V4.1-Flash is trained from scratch on a multimodal corpus containing 45 trillion tokens. Its sparse attention is trained at a 64K sequence length, with the context window extended to 1 million tokens after training on 34 trillion tokens.

The launch comes after the US government accused DeepSeek and five other Chinese AI companies of using AI model distillation to extract capabilities from US frontier models. 

DeepSeek is also moving toward a potential IPO in China. Reuters reported that the company has hired CITIC Securities to prepare for a listing on Shanghai’s STAR Market, with the process expected to begin this year. The timing, size, and valuation of the offering have not been finalised.