DeepSeek-V4.1-Flash debuts on qwen AI platform with enhanced features and cost efficiency

DeepSeek-V4.1-Flash, a new lightweight flagship model from DeepSeek, has been launched on the Qwen AI platform, alongside API services and a Token Plan. The model features a MoE architecture with 552B total parameters, employing a Causal-Encoder-Decoder asymmetric structure, and supports text and image understanding with a maximum context length of 1 million tokens. Industry experts noted that the model significantly improves over its predecessor in Agent and code benchmarks, achieving high throughput and low latency with lower activation parameters. The release emphasizes cost efficiency, with a new generation of cache compression reducing KV Cache demand for HBM to one-quarter and for SSD storage to one-eighth. Qwen offers time-based pricing: 1 yuan per million tokens during off-peak hours for input and 4 yuan per million tokens for output, with peak hours at 2 yuan and 8 yuan respectively. Alibaba Cloud BaiLian has adapted the model for deployment on both China and international sites, supporting multi-turn conversations and structured output. Experts believe the integration of 'large parameters, small activation, strong caching' into the Qwen ecosystem will reduce trial and error costs for long-context agents.