DeepSeek walks back its promise to keep AI prices discounted
Published: August 14, 19:15
Source: Engadget
DeepSeek has announced a new pricing structure for API access to its models, with changes going live on August 16 alongside the V4 Pro launch. Under the updated rates, the cost per million output tokens will vary by time of day. For DeepSeek V4 Pro, peak-period requests will be charged $3.96 per million output tokens, up from the current $0.87. Off-peak pricing will be $1.98 per million output tokens. The move comes as AI API pricing has become a major competitive battleground among model providers.
DeepSeek V4 Flash is also getting more expensive: peak usage will cost $1.32 per million output tokens, compared with $0.28 today, while off-peak requests will be priced at $0.66. The existing rates had been offered as a promotional deal that was scheduled to end on May 31. Earlier, DeepSeek had indicated it planned to maintain these reduced prices indefinitely.
The AI pricing competition
Despite the increase, DeepSeek models remain cheaper than several competitors. For example, the price per million output tokens:
- Kimi K3 by Moonshot: $15;
- OpenAI GPT-5.6 Sol: $30;
- OpenAI GPT-5.6 Luna: $1.20.
So while DeepSeek's new pricing is a sharp increase from its previous levels, the models still undercut some other options in terms of cost per output token.
Company representatives said the changes are meant to 'allocate resources more efficiently'.
Existing customers will need to factor in the new rates, as the shift could have implications for businesses that rely on these models commercially. It remains to be seen how the market responds and whether other AI providers adjust their own pricing strategies.