# DeepSeek V4

DeepSeek V4 is a language model from Deepseek with a 512K-token context window. Provider list price is $1.74 per million input tokens and $3.48 per million output; on Allocate you pay $1.86 and $3.72. It is a closed model served over API; the weights are not published.

## Pricing

| | Provider list | On Allocate |
| --- | --- | --- |
| Input, per M tokens | $1.74 | $1.86 |
| Output, per M tokens | $3.48 | $3.72 |
| Cached input, per M tokens | $0.20 | $0.21 |

Prices checked 2026-07-21.

## Facts

| Field | Value |
| --- | --- |
| Lab | Deepseek |
| Modality | Language |
| Context window | 512K tokens |
| License | Proprietary API |
| Open weights | No |
| Fine-tunable | No |
| Catalog id | deepseek/deepseek-v4-pro |

## What a real workload costs

Take 1,000,000 requests a month at 1,200 input and 350 output tokens each: 1,200M input and 350M output tokens. At list prices that is 1,200 × $1.74 + 350 × $3.48 = $3,306 a month. Billed on Allocate it is $3,537.

## Where it fits

DeepSeek’s frontier-class model with 512K tokens of context. At $1.74 per million input tokens and $3.48 out, it undercuts the frontier flagships on list price while staying close on reasoning-heavy work.

- Reasoning-heavy agents
- Long-document analysis
- Cost-sensitive production routes

## Common questions

### How much does DeepSeek V4 cost per million tokens?

Provider list price is $1.74 per million input tokens and $3.48 per million output tokens. On Allocate you pay $1.86 in and $3.72 out.

### What context window does DeepSeek V4 have?

512,000 tokens (512K). At roughly 0.75 words per token, that is about 384k words of English text per request.

### What does cached input cost on DeepSeek V4?

$0.20 per million tokens at list ($0.21 billed). Repeated prompt prefixes, such as a stable system prompt or tool definitions, bill at this rate instead of the full input price.

### Can I fine-tune DeepSeek V4?

No. DeepSeek V4 is a closed model served over API; the weights are not published. If you want a model you can train and own, start from an open-weights base in the catalog and fine-tune that.

### How do I call DeepSeek V4 on Allocate?

Send deepseek/deepseek-v4-pro in the model field of the OpenAI-compatible endpoint at api.allocate.network/v1, or point a route name (like prod/support-agent) at it so you can swap the model later without a deploy.

---

[HTML page](https://allocate.network/models/deepseek-v4) · [Machine-readable catalog](https://allocate.network/catalog.json)