DeepSeek V4-Pro
A 1.6T-parameter mixture-of-experts model with 49B active, released under MIT and callable through DeepSeek's own API with configurable reasoning effort. Since 16 August 2026 DeepSeek bills on a peak/off-peak schedule, so the hour you run a job changes the bill by exactly 2x — which makes it unusually well suited to scheduled batch work.
Pricing checked against provider documentation on . How we verify
deepseek-v4-pro reasoning open weights Input
$0.66
/ 1M tokens
Output
$1.98
/ 1M tokens
Context
1M
1,000,000 tokens
Max output
131K
tokens per request
Pricing caveat: Off-peak rate shown. Peak rates are exactly double ($1.32 input / $3.96 output) during 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday. Cache hits cost roughly 3% of a cache miss.
What it costs in practice
A workload of 10M input and 2M output tokens per month — roughly a moderately busy production assistant — runs $10.56/month on DeepSeek V4-Pro. The next cheaper option, Gemini 3.1 Flash-Lite , would cost $5.50.
Where DeepSeek V4-Pro fits
- Overnight batch reasoning jobs
- Self-hosted deployments needing MIT terms
- Cost-sensitive coding agents
- Chinese-language workloads
Specifications
- API model ID
- deepseek-v4-pro
- Provider
- DeepSeek
- Model family
- DeepSeek V4
- Released
- August 2026
- Knowledge cutoff
- March 2026
- Context window
- 1,000,000 tokens
- Max output
- 131,072 tokens
- Native reasoning
- Yes
- Relative latency
- medium
- Input modalities
- text
- Output modalities
- text
- Open weights
- MIT
- Cached input
- $0.02 / 1M tokens
- Batch discount
- —
Frequently asked
How much does DeepSeek V4-Pro cost?
DeepSeek V4-Pro costs $0.66 per million input tokens and $1.98 per million output tokens, with cached input at $0.02.
What is DeepSeek V4-Pro's context window?
DeepSeek V4-Pro accepts up to 1,000,000 tokens of context and can generate up to 131,072 output tokens per request.
Is DeepSeek V4-Pro a reasoning model?
Yes. DeepSeek V4-Pro performs native chain-of-thought before answering. Those thinking tokens are billed at the output rate, so budget above the sticker price.
What is DeepSeek V4-Pro best for?
Frontier-adjacent reasoning at open-weight prices. A 1.6T-parameter mixture-of-experts model with 49B active, released under MIT and callable through DeepSeek's own API with configurable reasoning effort. Since 16 August 2026 DeepSeek bills on a peak/off-peak schedule, so the hour you run a job changes the bill by exactly 2x — which makes it unusually well suited to scheduled batch work.