DeepSeek V4.1 Flash lands on Workers AI: 552B total parameters, only 8B active per token for input-heavy agents
Synopsis
DeepSeek V4.1 Flash 1017 is now available on Cloudflare Workers AI. The multimodal Mixture-of-Experts model has a 552B-parameter backbone but activates only 8B parameters per token during prefill and 16B during decode, which keeps input-heavy agentic workloads fast and low-cost. The 1017 release improves reasoning, coding, agentic tasks and visual benchmarks over the prior Flash version and completes tasks with shorter outputs, with image-plus-text requests, reasoning on by default, multi-turn function calling and a 1,048,576-token context window.
Interpretation
DeepSeek V4.1 Flash 1017 can now be called through Workers AI, via the Workers AI binding, the REST API, the OpenAI-compatible endpoint or AI Gateway. This is an update to DeepSeek V4.1 Flash, which was already offered on Workers AI. The update covers reasoning, coding, agentic tasks and visual benchmarks, and completes tasks with shorter outputs, which is the comparison between versions.
The model separates total parameters from per-token activation: the backbone holds 552B parameters, while prefill routes only 8B per token and decode routes 16B. A Mixture-of-Experts model selects a subset of parameters per token instead of running all of them. That activation pattern targets input-heavy agentic workloads, aiming to keep them fast and low-cost.
Callers can send images alongside text in one request, and reasoning is on by default at high effort, adjustable to low or max, or turned off via reasoning_effort. Reasoning no longer needs a separate switch to be used; it starts enabled. The model also supports function calling across multiple conversation turns and a full 1,048,576-token context window.
Perspective
The update is aimed at developers building on Cloudflare Workers AI and describes model specifications, capabilities and access paths. Use requires a Workers Paid plan or prepaid AI Gateway credits, with a default rate limit of 20 requests per minute under standard Workers AI billing and 50 requests per minute with prepaid AI Gateway credits. The source does not give specific benchmark scores or item-by-item before-and-after comparisons.
The source says the 1017 release improves reasoning, coding, agentic tasks and visual benchmarks but lists no specific metrics or comparison baselines. Reasoning is on by default at high effort and can be adjusted or turned off via reasoning_effort, so behavior depends on how callers set it. Rate limits differ by billing: 20 requests per minute under standard billing and 50 requests per minute with prepaid AI Gateway credits.