DeepSeek API pricing (2026): the per-token cost table
Short answer: Per million tokens, DeepSeek V4 Flash costs $0.14 input / $0.28 output — the cheapest frontier-adjacent model on this site — and V4 Pro costs $0.435 / $0.87. Both carry a 1M-token context window and cached-input pricing around 96–97% off. DeepSeek has signaled a price increase is coming, so verify current rates before budgeting production volume.
DeepSeek API model pricing (per 1M tokens)
| Model | Input | Output | Cached input | Context |
|---|---|---|---|---|
| V4 Flash (cheapest) | $0.14 | $0.28 | $0.0028 | 1M tokens |
| V4 Pro | $0.435 | $0.87 | $0.003625 | 1M tokens |
Prices are USD per million tokens (MTok). Current model builds as of this check: deepseek-v4-flash updated to DeepSeek-V4-Flash-0731, and deepseek-v4-pro updated to DeepSeek-V4-Pro-0813. Legacy V3.2 and R1 remain documented but are no longer the current offering.
Worked example
A request that sends 20,000 input tokens and generates 2,000 output tokens on V4 Flash:
- Input: 20,000 × $0.14 / 1,000,000 = $0.0028
- Output: 2,000 × $0.28 / 1,000,000 = $0.00056
- Total: ~$0.0034 per request
The same request on V4 Pro costs (20,000 × $0.435 + 2,000 × $0.87) / 1,000,000 = $0.0104 — about 3x V4 Flash, for meaningfully stronger reasoning.
How DeepSeek compares to the majors
DeepSeek is the price leader across the board in 2026. V4 Flash's $0.14 input rate undercuts GPT-5.6 Luna ($0.20), Gemini 3.5 Flash-Lite ($0.30), and is roughly 7x cheaper than Claude Haiku 4.5 ($1). Even V4 Pro at $0.435 beats every mid-tier and flagship offering from the other three providers on input price. The tradeoff isn't quality alone — it's data residency (processed under Chinese data law), tooling maturity, and uptime consistency. Full context in our DeepSeek review and Claude vs DeepSeek comparison.
Full cross-provider breakdown: cheapest LLM API, OpenAI pricing, Anthropic pricing, Gemini pricing.