← Models
Model APILLM
DeepSeek·MIT
DeepSeek V4.1 Flash
552B MoE · 8-16B active · Text and images
DeepSeek's V4.1 Flash: a 552B Mixture-of-Experts model with Compressed Sparse Attention and Engram conditional memory, activating about 8B parameters per token in prefill and 16B in decode. Text and image input, up to 1M tokens of context.
Model details
| Developed by | DeepSeek |
| Model family | DeepSeek-V4.1 |
| Architecture | Mixture-of-Experts, Compressed Sparse Attention |
| Total parameters | 552B backbone + 196B Engram conditional memory |
| Active parameters | ~8B per token in prefill, ~16B in decode |
| Context length | 1M tokens |
| Modalities | Text, Image → Text |
| Vocabulary | 129K |
| License | MIT |
On Autoloops
| Hardware | 4× B200 |
| API | OpenAI-compatible |
| Region | US |
| Pricing | $0.30 / $1.20 / $0.006 per 1M tokens (input / output / cached) |
| Model ID | deepseek-v4.1-flash |
Capabilities
- Thinking mode with reasoning_effort, or direct answers with thinking off
- Native tool calling and JSON-schema structured output
- Image understanding through DeepSeek's vision tower
- Follow-up turns of a conversation are served from the prefix cache and billed at the cached rate
- OpenAI-compatible chat completions via Autoloops
OpenAI-compatible chat completion
from openai import OpenAI
import os
client = OpenAI(
api_key=os.environ["AUTOLOOPS_API_KEY"],
base_url="https://api.autoloops.ai/v1",
)
stream = client.chat.completions.create(
model="deepseek-v4.1-flash",
messages=[{"role": "user", "content": "Tell me a fun fact about cats."}],
stream=True,
)
for chunk in stream:
print(chunk.choices[0].delta.content or "", end="")
