DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with 552B backbone parameters that natively processes images and text at up to 1M token context. Its Causal Encoder-Decoder architecture activates only 8B parameters during prefill and 16B during decode, cutting the KV cache footprint to roughly a quarter of DeepSeek-V4-Flash for cost-efficient agentic workloads.
DeepSeek V4.1 Flash can be customized with your data to improve responses. Fireworks uses LoRA to efficiently train and deploy your personalized model
DeepSeek V4.1 Flash is available via Fireworks' serverless API, where you pay per token. There are several ways to call the Fireworks API, including Fireworks' Python client, the REST API, or OpenAI's Python client.
On-demand deployments allow you to use DeepSeek V4.1 Flash on dedicated GPUs with Fireworks' high-performance serving stack with high reliability and no rate limits.
Run queries immediately, pay only for usage
DeepSeek V4.1 Flash is a multimodal Mixture-of-Experts model from DeepSeek, released September 10, 2026, that processes images and text natively and targets cost-efficient agentic workloads.
Code agents (terminal and repository tasks), visual agents over images and charts, and other cost-efficient agentic workloads. DeepSeek reports 90.6 on Terminal-Bench 2.1 and 74.2 on DeepSWE v1.1.
1,048,576 tokens, the max_position_embeddings value in config.json, reached by YaRN scaling (factor 16) from a 65,536-token base (original_max_position_embeddings).
Fireworks serves DeepSeek V4.1 Flash with a 1040k-token context window, available on both serverless and on-demand deployments.
Yes. The checkpoint is natively quantized to FP8 with block-wise scaling and dynamic activations, and its expert weights are FP4.
Fireworks allows generation up to the full 1040k-token context window. When max_tokens is not set, it will default to 2,048. DeepSeek recommends setting to at least 256K.
DeepSeek documents two limitations:
Yes to both. Streaming runs through the stream parameter, and the model page lists function calling as supported.
DeepSeek V4.1 Flash has 552B backbone parameters, activating 8B per token during prefill and 16B during decode.
No. Fine-tuning is not supported for DeepSeek V4.1 Flash on Fireworks, and no V4.1 variant offers it.
Fireworks bills input, cached input, and output tokens separately, priced at $0.30, $0.006, and $1.20 per 1M tokens.
Fireworks places DeepSeek V4.1 Flash in the Small tier for serverless rate limits. Limits adapt per account and model as usage grows, up to ceilings of 216M total prompt TPM, 54M uncached prompt TPM, and 2.16M generated TPM at spending Tier 3 and above. On-demand deployments carry no rate limits.
Fireworks announces serverless model deprecations in advance, following its serverless model lifecycle policy. For long-term version stability it recommends on-demand deployments.
MIT License, covering both the repository and the model weights. Commercial use, modification, and redistribution are permitted provided the copyright and permission notice are retained.
No. Fireworks operates zero data retention by default and does not log prompt or generation data for open models without opt-in. The Responses API is the exception, retaining data 30 days when store is true.
