DeepSeek V4.1 Flash Released: V4 Pro Retires September 14 with Automatic Routing

DeepSeek has announced DeepSeek V4.1 Flash. At the same time, the V4 Pro service is being retired on September 14, 2026 at 12:00 Beijing time — from that moment, every request sent to V4 Pro is automatically routed to V4.1 Flash and billed at V4.1 Flash rates. No code changes on your side.

Per the official note, after internal and external testing, V4.1 Flash has comprehensively surpassed V4 Pro across performance, cost, speed, and total time.

The model ID is deepseek-flash. It is already live on the TokenPAPA model marketplace, so one API key is all you need to call it today.

What is changing

The only thing to watch: routing is automatic, but the price changes. If your budget was built on the old V4 Pro rates, re-run the numbers before September 14.

Why V4.1 Flash "comprehensively surpasses" V4 Pro

The four dimensions in the official comparison:

In practice you notice two things: faster replies to the same prompt, and shorter waits on long tasks. If you run cost-sensitive batch jobs or latency-sensitive features, this upgrade moves in the right direction on both.

TokenPAPA pricing

Base rates for V4.1 Flash (deepseek-flash) on TokenPAPA:

Different account groups get different multipliers:

Two things stand out:

• Cache input is $0.006/1M, about 1/50 of the standard input rate. If your app reuses a system prompt or a fixed prefix, cached input becomes almost free.
• The group spread is significant: Enterprise at 50% off, Pro at 20% off, Starter at 10% off. The higher your tier and the more you use, the lower your unit price.

Cost example

Assume 100,000 requests per month, each around 1,500 tokens (1,000 input + 500 output):

If most of the input hits the cache, the input side drops by another order of magnitude.

What this means if you use V4 Pro

1. No code changes — keep calling V4 Pro; requests land on V4.1 Flash automatically;
2. Re-check your budget — run the cost comparison against the new V4.1 Flash rates;
3. Lean on caching — keep stable system prompts and fixed prefixes so they hit the cache and input drops to $0.006/1M;
4. Share findings — DeepSeek welcomes feedback from anyone comparing V4 Pro and V4.1 Flash in testing.

Quick start

from openai import OpenAI

client = OpenAI(
    api_key="your-tokenpapa-key",
    base_url="https://tokenpapa.ai/v1"
)

resp = client.chat.completions.create(
    model="deepseek-flash",   # DeepSeek V4.1 Flash
    messages=[{"role": "user", "content": "Explain the difference between V4.1 Flash and V4 Pro in one sentence."}]
)
print(resp.choices[0].message.content)

One API key covers DeepSeek V4.1 Flash, V4 Pro, plus GPT, Claude, Gemini, Qwen, Kimi, and 30+ more models — all through a single OpenAI-compatible endpoint. No Chinese phone number required, and switching models is a one-line model= change.

FAQ

Q: Will my V4 Pro calls break after the retirement? A: No. After 12:00 on September 14, V4 Pro requests are automatically routed to V4.1 Flash and billed at V4.1 Flash rates. No code changes are required.

Q: What is the model ID for DeepSeek V4.1 Flash? A: The model ID is deepseek-flash, now live on the TokenPAPA model marketplace.

Q: Is V4.1 Flash cheaper or more expensive than V4 Pro? A: Per the official results, V4.1 Flash beats V4 Pro on cost. Re-check your usage budget against the V4.1 Flash rates on TokenPAPA before September 14.

Q: How does the $0.006/1M cache input price work? A: When the prefix of your request (system prompt, few-shot examples, and so on) hits the cache, that portion of input is billed at $0.006/1M — about 1/50 of the standard input rate. Repetition-heavy apps save the most.

Get Started

1. Sign up at tokenpapa.ai;
2. Create an API key (no Chinese phone number needed);
3. Point your base_url to https://tokenpapa.ai/v1 and send your first request with the model ID deepseek-flash.

V4 Pro retires on September 14 at 12:00 — the migration is automatic, but the budget is worth checking early.