Skip to content
SI.info

SI Model · DeepSeek

DeepSeek V4 Pro

Published Verified 1 min read
On this page
  1. Timeline
  2. Key specs
  3. Pricing (per 1M tokens)
  4. Compare
DeepSeek V4 Pro specifications
Developer DeepSeek
Released August 12, 2026
Status Current
Inputs Text
Context window 1M tokens
API price (input) $1.32 / 1M tokens
API price (output) $3.96 / 1M tokens
Open weights Yes
Access DeepSeek app, DeepSeek API, Hugging Face (MIT licence)

Specs and list prices last verified October 8, 2026. Prices change often; check the provider before you buy.

Strengths

  • Open weights under the permissive MIT licence
  • Very low API prices with 50% off-peak discounts
  • 1M-token context

Best for

  • Teams that need to self-host or fine-tune a frontier-class model
  • Cost-sensitive API use
  • Research on open models

DeepSeek V4 Pro is the flagship of DeepSeek’s V4 generation, and one of the most capable open-weights models available.

Timeline

  • April 24, 2026: DeepSeek publishes preview weights for V4-Pro and V4-Flash on Hugging Face under the MIT licence.
  • August 12, 2026: V4 Pro leaves preview with full weights and a production API.

Key specs

  • Architecture: mixture of experts, reported at roughly 1.6–1.7 trillion total parameters
  • Context window: 1 million tokens (DeepSeek documents a 384K output limit for V4 Flash; it does not state one for V4 Pro)
  • Licence: MIT, which allows commercial use, modification and redistribution

Pricing (per 1M tokens)

Off-peakPeak
Input (cache hit)$0.022$0.044
Input (cache miss)$0.66$1.32
Output$1.98$3.96

Our tables show the peak (standard) price. A cheaper sibling, DeepSeek V4 Flash, costs $0.15–0.30 input and $0.60–1.20 output.

Compare

DeepSeek V4 Pro vs Qwen3.8-Max, the two largest Chinese open-weights flagships.

Frequently asked questions

How much does the DeepSeek V4 Pro API cost?

At peak times $1.32 per million input tokens (cache miss) and $3.96 per million output tokens; off-peak prices are half that ($0.66 and $1.98). Cached input is far cheaper. Peak hours are 01:00–04:00 and 06:00–10:00 UTC on weekdays.

Sources

  1. DeepSeek API pricing — DeepSeek
  2. DeepSeek V4 Pro exits preview with MIT-licensed 1.7T weights — AI Weekly

Written by

· Editor

Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.

How we research and fact-check

Free newsletter

Get The SI Brief

One short email a week: what changed in Super Intelligence, policy and models — and why it matters.

Free. One email a week. Sent via beehiiv, which counts opens and clicks. Unsubscribe anytime. Privacy policy