DeepSeek Reveals Full V4 Pro API Pricing and New Peak/Off-Peak Schedule [Updated]
DeepSeek has officially released its confirmed API pricing schedule for DeepSeek-V4-Pro and V4-Flash, introducing peak/off-peak rates and ultra-low prompt caching for enterprise AI agents.
DeepSeek posted a notice on its developer platform warning that prices for its API services would rise “significantly.” The company has now followed through with the full rollout of its official pricing schedule and the General Availability (GA) release of DeepSeek-V4-Pro.
[Update — August 16, 2026: Official Pricing and DeepSeek-V4-Pro GA Confirmed]
DeepSeek has officially ended the ambiguity surrounding its pricing adjustments by publishing its finalized API rate structure on the DeepSeek Official API Pricing Documentation, accompanied by the general availability rollout of DeepSeek-V4-Pro-0813.
The Final Numbers: V4 Pro and V4 Flash Rates
The updated structure establishes permanent rates for both the flagship DeepSeek-V4-Pro and the lightweight DeepSeek-V4-Flash, while introducing a formal peak/off-peak tiering mechanism effective August 16 at 16:00 UTC.
| Model Name | Architecture / Context | 1M Input (Cache Hit) | 1M Input (Standard / Cache Miss) | 1M Output Tokens |
|---|---|---|---|---|
| DeepSeek-V4-Flash | 284B MoE / 128K context | $0.0028 | $0.14 | $0.28 |
| DeepSeek-V4-Pro | 1.6T Total / 49B Active MoE (1M Context) | $0.003625 | $0.435 | $0.87 |
Peak window rates apply during peak global usage hours (typically 00:30 UTC to 16:30 UTC), while off-peak discounts provide up to a 50% reduction on input token processing.
What DeepSeek-V4-Pro Brings to Enterprise Deployments
Alongside the pricing finalization, DeepSeek confirmed the technical specifications for its flagship model:
- Mixture-of-Experts Scale: 1.6 trillion total parameters with only 49 billion activated parameters per token, maximizing inference throughput.
- Massive 1M Context Window: Full document and repository ingestion capabilities with ultra-low cache-hit pricing ($0.003625/M tokens), making continuous agentic memory loops economically viable.
- Thinking Mode Integration: Native support for both standard low-latency inference and extended reasoning chains for complex logic, math, and code generation.
Economic Impact on AI Agents & Autonomous Pipelines
For engineering teams building multi-agent systems, prompt caching is now the single most critical factor in cost optimization. Because agentic workflows frequently re-send massive context windows (system prompts, tool definitions, conversation history), DeepSeek’s aggressive context caching discount allows developers to execute multi-turn agent loops at a fraction of the cost of competing frontier APIs.
Historical Background: Where DeepSeek’s Prices Stood
DeepSeek’s initial V4-Flash model launched at $0.14 per million input tokens and $0.28 per million output tokens. On a weighted average per standardised task, that put the baseline cost at roughly $0.03 — a figure that forced the entire AI industry to re-evaluate API pricing assumptions.
That aggressive posture earned DeepSeek the moniker of creating a “death zone” in enterprise AI economics. According to tracking from DeepSeek Release Logs and OpenRouter benchmarks, the price-to-performance ratio established by Chinese open and commercial models altered global infrastructure spending patterns.
Frequently Asked Questions (FAQ)
What are the confirmed DeepSeek API prices as of August 16, 2026?
For DeepSeek-V4-Pro, prices are set at $0.435 per million standard input tokens ($0.003625 cache hit) and $0.87 per million output tokens. DeepSeek-V4-Flash costs $0.14 per million input tokens ($0.0028 cache hit) and $0.28 per million output tokens.
When does the new peak / off-peak pricing take effect?
The new peak and off-peak rate structure takes effect on August 16, 2026 at 16:00 UTC across the official DeepSeek API.
Is DeepSeek-V4-Pro available in general availability (GA)?
Yes. DeepSeek-V4-Pro has graduated from preview status and is fully available across the official Web, App, and developer API platforms.
How does prompt caching affect DeepSeek API costs?
Prompt caching reduces input token costs by up to 99%, charging only $0.003625 per 1M cached tokens on V4-Pro, significantly lowering the total cost of ownership for stateful AI agents and long-context applications.
