DeepSeek V4.1 Flash 1017 Lands on OpenRouter

DeepSeek V4.1 Flash 1017 is a dated snapshot of DeepSeek's V4.1 Flash model, served through OpenRouter under the deepseek-flash API name. The model runs on DeepSeek's new Causal Encoder-Decoder (CED) architecture — the first in the V4.1 line to do so — and is a sparse mixture-of-experts system with a 552B-parameter backbone. It activates 8B parameters on input and 16B on output, an asymmetric split designed to keep per-token compute low. Vision is baked in natively: visual and text embeddings were trained jointly from the start, rather than bolted on post-hoc as with the earlier V4 Flash Vision Exp. Compressed KV caching cuts cache memory to roughly one-quarter of the previous Flash generation, which DeepSeek says meaningfully lowers costs on agentic workloads. Target use cases include coding agents, computer-use tasks, and long-context analysis. DeepSeek claims V4.1 Flash outperforms its own V4 Pro on performance, speed, and task completion time — positioning it as the affordable tier of the V4.1 family. Priced at $0.10 input / $0.50 output per 1M tokens with a 1M-token context window. Not available via EU in-region routing.