DeepSeek has released V4.1-Flash, a new open-weight flagship model designed for extreme inference efficiency and low cost. The model utilizes a causal encoder-decoder architecture to significantly reduce active compute and KV/cache costs.

  • It scores 40 on the Artificial Analysis Intelligence Index, surpassing DeepSeek V4 Pro 0813 while costing $0.30 per 1M input tokens.
  • The architecture features 763B total parameters with only 8B active for input and 16B active for output.
  • It supports a 1M-token context window and accepts both text and image inputs under an MIT license.
  • Independent evaluations show it ties GPT-6 Astra on AutomationBench-AA at 69% and matches GPT-5.6 Sol on AA-LCR v1.1 at 84%.
  • Users report running the model locally on commodity hardware, achieving high throughput via SSD streaming and offloading.

The release positions V4.1-Flash as a cost-effective alternative to proprietary models, offering strong performance metrics at a fraction of the price.