DeepSeek may be preparing customers for a substantial API price increase. Clients have reportedly received advance notices, but the company has not publicly confirmed the size of the increase or when it could take effect.
That leaves developers with a planning problem rather than a firm new bill. Teams that chose DeepSeek primarily for price should review their usage before making any commitments, while avoiding a rushed migration based on details that could still change.
How the tentative pricing compares
The available pricing figures have not been independently confirmed. They put DeepSeek’s V4 Flash model at $0.14 per million input tokens and $0.28 per million output tokens. The comparison figure for Google’s Gemini 2.5 Flash-Lite is $0.10 for input and $0.40 for output.
| Model | Input per million tokens | Output per million tokens |
|---|---|---|
| DeepSeek V4 Flash | $0.14 | $0.28 |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 |
At those unconfirmed rates, Gemini would cost less for input, while DeepSeek would remain cheaper for generated output. That distinction matters: an application that processes long documents can have a different cost profile from a chatbot producing lengthy answers.
The practical comparison is not the headline price alone. Buyers should run their own input-to-output ratio through both rate cards and account for performance, reliability, latency, and migration work before choosing an AI API provider.
Peak-hour pricing adds another variable
A separate, also-unconfirmed pricing change has been described for peak-hour API use. Under that arrangement, rates would effectively double during certain daily windows as DeepSeek attempts to reduce pressure on its servers. It is unclear how that policy would interact with the broader increase being discussed.
For workloads that can be queued or scheduled, moving non-urgent jobs outside expensive windows may reduce the impact. Real-time products have less flexibility and could face a sharper increase in operating costs.
What API customers should do
Until DeepSeek publishes a firm rate card and effective date, customers can prepare without overreacting:
- Record a baseline of monthly input and output token use.
- Model several possible price increases instead of relying on one forecast.
- Check whether workloads can move away from peak-rate periods.
- Benchmark alternatives from Google, OpenAI, and Anthropic using the same workload.
- Include engineering and migration costs in the final comparison.
The reported DeepSeek price hike may ultimately change the provider’s value proposition. For now, the most useful response is to understand which parts of a workload drive the bill and how easily they could move elsewhere.