DeepSeek V4.1-Flash, text and images at 384K context
DeepSeek's V4.1-Flash is a multimodal mixture-of-experts with 552B backbone parameters, MIT-licensed. This endpoint serves its October 17 update through chat completions or the Anthropic Messages API.
- 552B MoE
- Image input
- 384K context
- MIT license
Three tiers, per token.
Per 1M tokens
| Tier | Input | Cached | Output |
|---|---|---|---|
| Sync | $0.30 | $0.006 | $1.20 |
| AsyncEarly access | $0.24 | $0.0048 | $0.96 |
Async is 20% off Sync.
openrelay/deepseek-v4.1-flashFull catalog on the inference pricing page. Add a card and make a deposit to start.
from openai import OpenAI
client = OpenAI(
base_url="https://inference.openrelay.inc/v1",
api_key="or_••••••••", # same SDK, new base URL
)
resp = client.chat.completions.create(
model="openrelay/deepseek-v4.1-flash",
messages=[{"role": "user", "content": "What is the weather in Lisbon?"}],
tools=[{"type": "function", "function": {
"name": "get_weather",
"parameters": {"type": "object",
"properties": {"city": {"type": "string"}}},
}}],
)
print(resp.choices[0].message.tool_calls)The spec sheet.
What the model is, what it takes in, and the surface it serves on.
- Model id
- openrelay/deepseek-v4.1-flash
- Parameters
- 552B backbone (mixture-of-experts), 8B active per token while reading the prompt and 16B while generating
- Architecture
- 384 routed experts with 6 active per token plus one shared, and 196B of Engram memory tables read by token lookup
- Checkpoint
- DeepSeek-V4.1-Flash-1017, the October 17 update, served at its native FP8 with FP4 routed experts
- Modality
- Text and image in, text out
- Context
- 384K tokens (393,216) on this endpoint, prompt and completion combined. The model supports up to 1M.
- Endpoint
- POST /v1/chat/completions or POST /v1/messages, with tool calling and reasoning
- License
- MIT, open weights
Where DeepSeek V4.1-Flash earns its place.
What this model is actually for, versus the rest of the catalog.
8B active on the prompt
DeepSeek's encoder-decoder design activates 8B parameters per prompt token and 16B per generated token, aimed at agent workloads that read far more than they write.
Vision in the same weights
DeepSeek trained the vision encoder with the language model from the start of pre-training. Images go in as message parts on the same model id.
OpenAI or Anthropic SDK
Chat completions and the Anthropic Messages API both serve this model id, so either SDK calls the same model.
DeepSeek V4.1-Flash, answered.
Is there a DeepSeek V4.1-Flash API?
Yes: openrelay/deepseek-v4.1-flash, on the OpenAI-compatible chat completions endpoint at inference.openrelay.inc/v1 and on the Anthropic Messages API. The OpenAI SDK works with a new base URL and an OpenRelay key.
What does DeepSeek V4.1-Flash cost?
The rate on this page is read from the catalog when the page renders. Input and output tokens bill separately, a cached prompt prefix bills at a lower input rate, and reasoning counts as output.
How much context can I send?
393,216 tokens (384K) on this endpoint, prompt and completion combined. DeepSeek trained V4.1-Flash for up to 1M; this deployment runs a shorter window.
Which checkpoint does it serve?
DeepSeek-V4.1-Flash-1017, the October 17 update to V4.1-Flash. DeepSeek reports gains in reasoning, coding, agent tasks and visual benchmarks, with shorter outputs.
First request in five lines.
Point the OpenAI SDK at inference.openrelay.inc/v1 and run DeepSeek V4.1-Flash per token. No contract, no minimums.