Overview
Run DeepSeek V4.1 Flash on Crusoe with serverless inference. A 552B open-weight model with image input and a 1M-token context, built for coding and agentic workloads.
Explore additional models
Frequently
asked questions
DeepSeek V4.1 Flash is a 552B-parameter mixture-of-experts model released in September 2026, and the first model built on DeepSeek’s new Causal Encoder-Decoder architecture. It activates about 8B parameters when reading input and about 16B when generating output, accepts text and image input natively, and supports a 1M-token context. It ships under the MIT license and is available for Serverless Inference in Crusoe Intelligence Foundry.
DeepSeek V4.1 Flash is the newer generation, with a new architecture, native image input, and a much smaller KV cache, which makes long-context and agent workloads cheaper to run. DeepSeek V4 Flash 0731 is text-only on the previous V4 architecture. On Crusoe, V4.1 Flash is available for Serverless Inference, while V4 Flash 0731 is available for Self-Serve Deployments and Serverless Fine-Tuning.
DeepSeek V4.1 Flash on Serverless Inference is billed per token, with separate rates for uncached input, cached input, and output tokens, and no minimum commitment. Cached input is billed at a fraction of the fresh input rate, which matters for agent loops and long documents that resend the same context. Current rates are shown in the pricing module above.
DeepSeek reports V4.1 Flash outperforming its previous flagship, V4 Pro, on coding, math, and agent benchmarks, while running faster and at a lower cost. On DeepSWE, a software engineering benchmark, it scores 74.2 by DeepSeek’s own measurement, level with leading closed models.
DeepSeek V4.1 Flash is available for Serverless Inference in Crusoe Intelligence Foundry, with no infrastructure setup required. It is not currently offered for Self-Serve Deployments or Serverless Fine-Tuning. For a DeepSeek model you can fine-tune or run on a dedicated GPU, use DeepSeek V4 Flash 0731. For an SLA-backed dedicated endpoint, contact us about Tailored Deployments.
MemoryAlloy technology is a proprietary memory fabric in Crusoe’s inference engine that shares the KV cache across nodes, with cache-aware routing that sends each request to the node most likely to have its context ready. Paired with V4.1 Flash’s compact KV cache, repeated long-context calls like agent loops and large codebases skip redundant prefill, so responses start faster and reused context is billed at the lower cached-input rate.


%204.png)
