Skip to content
CheapestInference

DeepSeek V4.1 Flash API — unlimited & flat-rate access

DeepSeek V4.1 Flash is DeepSeek’s new-generation fast model — a 552B-parameter mixture-of-experts model (8B active in prefill, 16B in decode) built on a new causal encoder–decoder architecture, with native image input and a 1M-token context. CheapestInference serves it in the Core Pool through an OpenAI- and Anthropic-compatible API on unlimited, flat-rate time-block subscriptions — so your cost does not scale with tokens.

Quick facts

Section titled “Quick facts”
ModelDeepSeek V4.1 Flash
VersionV4.1-Flash — new generation, live since 2026-09-10, what changed
Model IDdeepseek-v4.1-flash
Context window1M tokens
VisionYes — image input on the same endpoint
Max request size8 MB per request
Open weightsYes — deepseek-ai/DeepSeek-V4.1-Flash, MIT license
Per-token price elsewhere$0.15 in / $0.60 out per 1M off-peak — reference only
PoolCore Pool (with MiMo v2.5)
Endpoints/v1/chat/completions (OpenAI), /anthropic/v1/messages (Anthropic)
Pricing$17.99/mo per 8-hour daily time block ($15.29/mo with annual billing), up to full 24/7

Call DeepSeek V4.1 Flash

Section titled “Call DeepSeek V4.1 Flash”
from openai import OpenAI


client = OpenAI(
    base_url="https://api.cheapestinference.com/v1",
    api_key="sk-..."  # your subscriber key
)


response = client.chat.completions.create(
    model="deepseek-v4.1-flash",
    messages=[{"role": "user", "content": "Summarize this document..."}],
)
Terminal window
curl https://api.cheapestinference.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-v4.1-flash", "messages": [{"role": "user", "content": "Hello"}]}'

Image input uses the standard formats on the same endpoints: image_url content parts on /v1/chat/completions, or image blocks on /anthropic/v1/messages — in user messages, JPEG, PNG, GIF or WebP, within the 8 MB per-request limit.

Why flat-rate DeepSeek V4.1 Flash

Section titled “Why flat-rate DeepSeek V4.1 Flash”

DeepSeek V4.1 Flash is built for speed at frontier-ish quality: DeepSeek reports it ahead of its own V4 Pro on performance, cost, speed and total time to finish a task, with a KV cache that needs a quarter of the HBM and an eighth of the SSD of the previous generation. That profile suits high-volume pipelines, million-token document work, screenshots and diagrams alongside text, and long agent conversations. On CheapestInference it is billed at a flat monthly rate, not per token, so bulk workloads have a fixed, predictable cost — a Core Pool time block covers unlimited DeepSeek V4.1 Flash usage (plus MiMo v2.5) during your reserved hours. It works in any OpenAI-compatible client.

Common questions

Section titled “Common questions”

Is there a cheap unlimited DeepSeek API? Yes. A CheapestInference Core Pool subscription gives unlimited DeepSeek V4.1 Flash usage — no token counting, no overage charges during your reserved daily 8-hour blocks — for a flat $17.99/month per block ($15.29/mo with annual billing). It is an OpenAI- and Anthropic-compatible drop-in: point your SDK at https://api.cheapestinference.com/v1 and use model id deepseek-v4.1-flash.

What happened to deepseek-v4-flash? The Core Pool’s DeepSeek slot moved to V4.1 Flash. Requests with the old id deepseek-v4-flash are still accepted and served by V4.1 Flash until 2026-10-10; after that date the id returns an invalid-model error. Migration is one line — change the model id to deepseek-v4.1-flash. Nothing else changes: same key, same endpoints, same subscription. See DeepSeek V4 Flash — retired.

How much does DeepSeek V4.1 Flash cost? From $17.99/month ($15.29/mo with annual billing). You reserve one or more 8-hour daily time blocks (up to full 24/7) and use DeepSeek V4.1 Flash with no token caps — billed at a flat monthly fee, not per token.

Can DeepSeek V4.1 Flash read images? Yes — it was trained with vision from pre-training, so image input is native rather than bolted on. Send image_url content parts (OpenAI format) or image blocks (Anthropic format) in user messages: JPEG, PNG, GIF and WebP, inside the same 8 MB request budget as everything else.

How large can a request be? Each request to the Core Pool accepts up to 8 MB of payload — messages, history and attachments combined. Larger requests are rejected with a context_length_exceeded error; agent clients such as Claude Code compact the conversation automatically and retry. See Request size limits.

How does DeepSeek V4.1 Flash compare to Claude Opus? On DeepSeek’s own agent benchmarks it edges ahead of Claude Opus 5.0: Terminal-Bench 2.1 90.6 vs 89.1, DeepSWE v1.1 74.2 vs 74.0 resolved, AutomationBench 54.8 vs 50.3 pass@1 — all three run by DeepSeek on its own harness, with no independent verification, so read them as vendor numbers. Artificial Analysis has not published an Intelligence Index score for V4.1 Flash yet. Full comparison.

Is DeepSeek V4.1 Flash OpenAI-compatible? Yes. Use model id deepseek-v4.1-flash against https://api.cheapestinference.com/v1 with the OpenAI SDK, or the /anthropic endpoint with the Anthropic SDK.

Section titled “Related”