DeepSeek V4.1-Flash
DeepSeek's latest open-weight Flash model: cheaper per token than V4-Flash 0731, with agentic results DeepSeek reports ahead of V4-Pro.
Architecture
MoE
552B total / 8B active (in)
Context Window
1M
Tokens
Provider
DeepSeek
Open Weights
Quantization
FP4 / FP8
Released 10 Sept 2026
A Leaner Flash, Built for Agents
DeepSeek-V4.1-Flash is DeepSeek's September 2026 release: a 552B-parameter MoE with 8B active parameters per input token and 16B per output token, native vision, a 1M-token context window, and a KV cache about a quarter the size of V4-Flash per token.
It supports thinking and non-thinking modes with a continuous reasoning-effort setting, DSML tool calling and JSON output. Available in FP4 and FP8 quantizations. Released 10 Sept 2026.
552B MoE · 8B active
Best for agentic coding and long-context work
Agentic Coding
DSML tool calling and JSON output for coding agents that plan, edit, and verify across many steps.
Multi-Tool Workflows
Reliable tool selection and structured outputs for agents orchestrating several services at once.
Long-Context Document Work
1M-token context handles whole codebases, contracts, and research corpora in a single call.
Thinking & Non-Thinking Modes
Continuous reasoning-effort setting lets you trade depth against cost per request.
Flexible Pricing Tiers
Choose the optimal balance of speed and cost for your workflow. Prices are per 1M tokens.
| Tier | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Batch (24H) | $0.08 | $0.30 |
| Async | $0.12 | $0.48 |
| Realtime | $0.15 | $0.60 |
Context window natively supported up to 1M tokens. An Artificial Analysis Intelligence Index score is not yet published for this model.
Start Building in Minutes
DeepSeek V4.1-Flash is accessible via OpenAI-compatible endpoints.
from openai import OpenAI
client = OpenAI(
api_key="your-api-key-here",
base_url="https://api.doubleword.ai/v1"
)
# Agentic, long-context work with DeepSeek V4.1-Flash (1M tokens)
response = client.chat.completions.create(
model="deepseek-ai/DeepSeek-V4.1-Flash",
messages=[
{"role": "user", "content": "Analyze this repository and propose a refactor plan."}
],
)
print(response.choices[0].message.content)💡 Pro Tip
The smaller KV cache makes V4.1-Flash especially cheap on long agent traces — pair it with the Async or Batch tier for bulk pipelines.