DeepSeek: DeepSeek V4.1 Flash
Overview
DeepSeek's current Flash model (DeepSeek-V4.1-Flash), a 552B-parameter MoE with native image understanding and thinking and non-thinking modes. Weights are published on Hugging Face.
- Provider
DeepSeek
- Modality
- Text and image
- Input
- text, image
- Output
- text
- Max output
- 384,000 tokens
- Weights
- Open weights
- Supported parameters
- tools, response_format, reasoningProvider's model page, checked .
- Distillable
- YesLicence or terms, checked .
- Tags
- open-weights, reasoning, coding, vision, long-context, low-cost
Data policy
- Prompts go to
- DeepSeek (People's Republic of China)
- Retention
- Privacy policy: personal data stored in the PRC and kept as long as necessary to provide the service, may be used to train models, with a right to opt out. No API-specific retention period or ZDR offering published
- Zero data retention
- Not stated by the provider
- Region
- China (PRC)
- Policy
- DeepSeek data policy
Summarised from the provider's published terms. The provider's own policy is what applies.
Where you can run it
How DeepSeek V4.1 Flash can be deployed, bought and reached through Safeguard.
- Deployment
- Private cloudOn-prem
- Procurement
- Not published yet
- Inference regions
- Not published yet
On-prem deployments run wherever you install them.
Pricing
| Provider | Input / 1M tokens | Output / 1M tokens | Context |
|---|---|---|---|
| DeepSeek | $0.30 | $1.20 | 1M |
- Notes
- Peak-hour list price (cache miss). Off-peak is half: $0.15 in / $0.60 out. Peak hours 01:00-04:00 and 06:00-10:00 UTC Mon-Fri excluding Chinese public holidays. Cache-hit input $0.006 peak.
- Source
- DeepSeek pricing , last verified November 22, 2026
No markup on model usage: you pay the provider's list price.
API
OpenAI-compatible. Your API key and access to this model are set up for your workspace when you choose Use with Safeguard.
from openai import OpenAI
client = OpenAI(
base_url="https://api.safeguard.sh/gpt/v1",
api_key="SAFEGUARD_API_KEY",
)
response = client.chat.completions.create(
model="deepseek/deepseek-flash",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)Base URL https://api.safeguard.sh/gpt/v1, model deepseek/deepseek-flash.
Self-healing security runs on Safeguard.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.