Skip to content
AI Model Radar

DeepSeek V4.1 Flash 1017

deepseek-ai/DeepSeek-V4.1-Flash-1017
Open Source · chat · open-weights
Open Alert me on changes
Context
1M
Max output
384K
Weights
Open
API $/1M
$0.15 / $0.60
Modalities
text · image
Released
17 Oct 2026
License: mit · deepseek-ai/DeepSeek-V4.1-Flash-1017
Download image Share on X Share on LinkedIn
AI summary
● machine-written

DeepSeek V4.1 Flash 1017: sparse mixture-of-experts model with 1M context window

DeepSeek-V4.1-Flash-1017 is a sparse mixture-of-experts model built on the company's Causal Encoder-Decoder architecture, activating 8B parameters on input and 16B on output from a 552B-parameter backbone. It supports multimodal input (text and image) with a 1M-token context window and is positioned as the cost-efficient tier of the V4.1 family. The model is available under MIT license as open-weights on Hugging Face.

What's new
  • Native image understanding with visual and text embeddings trained jointly from pre-training start
  • Compressed KV caching reduces cache memory to roughly a quarter of previous Flash generation
  • Asymmetric sparse architecture keeps per-token compute low relative to total model size
  • Supports both non-thinking and thinking modes via API
Best for
Coding and computer-use agentsLong-context analysis and multi-step tasksTerminal and agentic workloadsCost-sensitive applications requiring multimodal input
Sources

Source: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017