Open weights · MIT
DeepSeek V4.1 Flash
DeepSeek's smallest model on its new Causal Encoder-Decoder architecture, a 552B MoE activating 8B parameters on input and 16B on output, with native image input and MIT-licensed weights.
Key Specifications
- Context window
- 1M tokens
- Max output
- 384K tokens
- Released
- Parameters
- 552B, 16B active
- Inputs
- Text, image
- Capabilities Show details
- Reasoning Thinks before it answers, always on or as a switchable mode. Function calling Connect to external tools, APIs, and systems. Structured output Return responses in structured formats like JSON.
Hosted API pricing
| Provider | Input / 1M tokens | Output / 1M tokens | Cached input / 1M | Cost at 10M in + 2M out | |
|---|---|---|---|---|---|
|
DeepSeek
Creator
|
$0.30 | $1.20 | $0.01 | $5.40 | View |
Novita
|
$0.30 | $1.20 | $0.01 | $5.40 | View |
Heads up: Base-tier, on-demand rates per 1M tokens; cached, batch and long-context tiers excluded. A provider may serve a shorter context or a quantized build than the creator's release. Verify before provisioning. More on how we price.
Estimated cost to self-host
DeepSeek V4.1 Flash needs about 314 GB of GPU memory at 4-bit with 32K context. The cheapest rental that fits is 8× RTX A6000 at about $3,168 a month.
That costs the same as roughly 7B tokens a month on DeepSeek's API. Self-hosting is more expensive below that volume.
| Precision | Memory | Cheapest fit (32K context) | Cost / mo | Break-even vs API |
|---|---|---|---|---|
| 4-bitINT4 / FP4 |
314 GB
|
$3,168
|
7B tokens / mo
|
|
| 8-bitFP8 / INT8 |
562 GB
|
$10,138
|
23B tokens / mo
|
|
| 16-bitFP16 / BF16 |
1,114 GB
|
$34,618
|
77B tokens / mo
|
Estimates based on median on-demand rates for Nvidia GPUs. Memory is weights plus KV cache for one request, using FP8 KV cache where supported and FP16/BF16 otherwise. KV cache assumed at 250 KB per token, a conservative default. Break-even assumes a 5:1 input-to-output ratio. No guarantee of runtime support, usable performance, or that a matching quantized build exists. How we estimate costs.
Similarly priced models
The models nearest DeepSeek V4.1 Flash by blended rate, each at its own cheapest provider.
| Model | Blended / 1M | Input / 1M | Output / 1M | Context | Cutoff | vs DeepSeek V4.1 Flash |
|---|---|---|---|---|---|---|
|
DeepSeek V3.1 Terminus
DeepSeek
|
$0.3917 | $0.27 | $1.00 | 131K | −13% | |
Codestral
Mistral
|
$0.40 | $0.30 | $0.90 | 128K | −11% | |
|
DeepSeek V3
DeepSeek
|
$0.4117 | $0.27 | $1.12 | 128K | −9% | |
Qwen3.7-Plus
Alibaba Cloud
|
$0.4167 | $0.28 | $1.10 | 1M | −7% | |
|
Qwen3-VL-Plus
Alibaba Cloud
|
$0.4333 | $0.20 | $1.60 | 262K | −4% | |
|
DeepSeek V4.1 Flash
This model
DeepSeek
|
$0.45 | $0.30 | $1.20 | 1M | ||
MiniMax-M2.5
MiniMax
|
$0.45 | $0.30 | $1.20 | 205K | 0% | |
Gemini 3.1 Flash-Lite
Google Cloud
|
$0.4583 | $0.25 | $1.50 | 1M | Jan 2025 | +2% |
|
Qwen3.6-Flash
Alibaba Cloud
|
$0.4583 | $0.25 | $1.50 | 1M | +2% | |
|
Qwen3-VL-235B-A22B Instruct
Alibaba Cloud
|
$0.50 | $0.30 | $1.50 | 262K | +11% | |
|
DeepSeek V4 Pro
DeepSeek
|
$0.5233 | $0.45 | $0.89 | 1M | +16% |
Prices are USD per 1M tokens at each model's cheapest listed provider. Blended is the cost of 10M input plus 2M output tokens, spread over the 12M.
Frequently Asked Questions
What is DeepSeek V4.1 Flash good for?
Long-context agent and coding work, and tasks that mix images with text. MIT-licensed weights, far lighter to self-host than V4 Pro, though the 552B backbone still takes a multi-GPU node.
When is DeepSeek V4.1 Flash not a good fit?
Audio and video work: it takes text and images in, and returns text only. Self-hosting still takes a multi-GPU node, and DeepSeek's own API charges the list rate at peak hours, so the price depends on the time of day.
What is the cheapest way to run DeepSeek V4.1 Flash?
Hosted, unless you push serious volume. DeepSeek charges $0.30 in / $1.20 out per 1M tokens. The cheapest rental that fits is 8x RTX A6000 at $3,168 a month, which costs the same as about 7B tokens a month on that API.
How does DeepSeek V4.1 Flash compare with DeepSeek V4 Flash?
DeepSeek says V4.1 Flash moves the Flash line onto a Causal Encoder-Decoder architecture that activates fewer parameters on input than on output, adds native image understanding, and compresses the KV cache, which it says cuts the cache-hit charges that dominate agent costs. Same 1M-token context. DeepSeek V4.1 Flash costs 275% more per 1M input tokens and 380% more per 1M output tokens than DeepSeek V4 Flash, at each model's cheapest listed provider. It came out 5 months later, has more parameters (552B with 16B active against 284B with 13B active) and takes image input, which DeepSeek V4 Flash does not.
Can I self-host DeepSeek V4.1 Flash?
Yes. The weights are MIT licensed. At 4-bit it needs about 314 GB of GPU memory, which starts at roughly $3,168 a month on the cheapest rental that fits.
More from DeepSeek
| Model | Context | Input / 1M | Output / 1M |
|---|---|---|---|
|
DeepSeek V3
Open weights
|
128K | $0.27 | $1.12 |
|
DeepSeek V3.1
Open weights
|
131K | $0.27 | $1.00 |
|
DeepSeek V3.1 Terminus
Open weights
|
131K | $0.27 | $1.00 |
|
DeepSeek V4 Pro
Open weights
|
1M | $0.45 | $0.89 |
|
DeepSeek V3 Turbo
Open weights
|
64K | $0.40 | $1.30 |
|
DeepSeek V4 Flash Vision Exp
|
1M | $0.44 | $1.32 |
|
DeepSeek V3.2
Open weights
|
131K | $0.25 | $0.80 |
|
DeepSeek R1 Distill Llama 70B
Open weights
|
131K | $0.80 | $0.80 |