DeepSeek API Pricing (October 2026): Flash vs V4 Pro, Peak vs Off-Peak

Quick answer: DeepSeek has two API tiers. deepseek-flash (DeepSeek-V4.1-Flash-1017) is the low-cost default. deepseek-v4-pro costs several times more per token. Both are cheaper off-peak, which includes all weekend, and cache hits cost a small fraction of a cache miss. Current rates are on the official DeepSeek pricing page.

Which DeepSeek model tier should you use?

Tier	Best for	Key differences	Official pricing
deepseek-flash	Most coding, chat and agent work	Cheapest tier; 1M context, up to 384K output	DeepSeek pricing
deepseek-v4-pro	Hardest reasoning tasks	Several times Flash’s per-token rate; same context limits	DeepSeek pricing
Search / Fetch tools	Web-grounded answers	Billed per 1,000 calls, with billing starting November 1, 2026	DeepSeek pricing

Each tier has three token rates: cache-hit input (cheapest), cache-miss input, and output (most expensive).

What changed with V4.1-Flash-1017?

DeepSeek launched it on October 17 with “stronger benchmark performance, faster task completion, and shorter outputs than the previous version” (@deepseek_ai). Shorter outputs lower the bill because output is the most expensive token type. The pricing page says the old deepseek-v4-flash name is now served by the 1017 model, so existing code moves over without changes.

When are DeepSeek peak and off-peak hours?

Peak pricing applies 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays. All other hours, and all of Saturday and Sunday, are off-peak. On the current price list, peak rates are double the off-peak rates, as developers on X also noted (@anya_yfa7).

Your time zone	Peak windows (weekdays)
UTC	01:00-04:00, 06:00-10:00
US Eastern (EDT, UTC-4)	21:00-00:00 (Sun-Thu evening), 02:00-06:00
US Pacific (PDT, UTC-7)	18:00-21:00 and 23:00-03:00 (Sun-Thu evening into the next day)
Central Europe (CEST, UTC+2)	03:00-06:00, 08:00-12:00

These conversions use daylight-saving offsets valid in October 2026; they shift by an hour when clocks change.

Is DeepSeek cheaper than OpenAI or Claude?

At list prices, deepseek-flash sits in the same budget band as GPT-6 Luna and Claude Haiku 5.5, and far below mid-tier models like GPT-6.1 Sol or Claude Sonnet 5.5. Quality and speed differ by task, so compare cost per finished task. Check each official page: DeepSeek, OpenAI, Anthropic. Our guides: OpenAI API pricing, Claude API pricing.

How do I switch existing code to DeepSeek?

1. Create a key on the DeepSeek platform and top up your balance.
2. Point your OpenAI SDK at https://api.deepseek.com, or your Anthropic SDK at https://api.deepseek.com/anthropic.
3. Set model to deepseek-flash or deepseek-v4-pro.

from openai import OpenAI
client = OpenAI(api_key="YOUR_DEEPSEEK_KEY", base_url="https://api.deepseek.com")
r = client.chat.completions.create(model="deepseek-flash",
    messages=[{"role": "user", "content": "Hello"}])

How can I keep DeepSeek costs lowest?

• Schedule batch jobs off-peak. The same job costs half as much on a weekend at current rates.
• Reuse prompt prefixes. A cache hit costs a small fraction of a miss.
• Cap output length. With up to 384K output tokens allowed, set max_tokens so runaway answers don’t run up the bill.

Hit an error while switching? See how to fix 401, 429 and quota errors. More guides are in the AI API hub.