Skip to content

DeepSeek settings guide

DeepSeek (V4) · Reasoning Model · Open Source

Chat / LLMFree tierAPIPopular
Token burn
Low burn
Paid plans from
API from $0.15/1M input tokens

Token-efficient; free tiers go a long way. Last reviewed September 2026; plans and model names change often, so check the tool's own pricing page before you buy.

1

The API has two models: deepseek-flash (DeepSeek-V4.1-Flash) for fast, cheap work and deepseek-v4-pro for harder tasks; both have a 1M-token context (api-docs.deepseek.com, September 2026).

2

The API is OpenAI-compatible: set base_url to https://api.deepseek.com and reuse the OpenAI SDK.

3

Cache repeated prompt prefixes: cache-hit input is billed at a small fraction of cache-miss input.

4

Off-peak rates are half the peak rates. Peak is 01:00–04:00 and 06:00–10:00 UTC on weekdays, so schedule batch jobs outside those windows.

5

Use flash for drafting and classification; move to v4-pro only where quality measurably improves.

6

Always set max_tokens on API calls.

7

Check your data rules before sending customer data to any external API, including this one.

API cost estimate

Monthly cost at list prices (September 2026), before and after a token cut you choose.

Now
$0.25
₹21
After a 25% cut
$0.19
₹16
Saved per month
$0.06
₹5

Assumes about 600 input and 200 output tokens per request (low burn). Excludes caching discounts, search or request fees and taxes. Check the vendor's pricing page before budgeting.

Reveaxa guide reference: RVX-T-deepseek