DeepSeek Adds Peak Surcharges to AI API Services

Chinese artificial intelligence firm DeepSeek is introducing peak hour surcharges for its API services. This move departs from a domestic price competition the company itself initiated, where competitors have engaged in aggressive price reductions to secure market share. The changes affect access to its V4 AI models.

Credit: UNSPLASH

The Hangzhou based organisation will implement a double price increase for access to its V4 AI models. This will occur via its application programming interface during specific peak periods. These times are designated as 9 a.m. to noon and 2 p.m. to 6 p.m. Beijing time. This information was detailed in an email distributed to service subscribers, which the South China Morning Post reviewed.

And the firm stated the price adjustment is necessary for resource allocation. DeepSeek communicated that the intention is to improve resource distribution and to enhance the stability of its services. This reasoning was provided within the same email correspondence.

Under the recently introduced pricing structure, the cost for DeepSeek's V4 Pro during peak hours will see an increment. It will rise to 12 yuan, or approximately USD 1.77, for every million output tokens. This compares to the previous standard rate of 6 yuan applicable during non peak periods. The company confirmed that these new prices will apply across all V4 models. This was announced in a separate public notice found on its official website.

But developers and enterprises generally understand that AI companies typically sell access to their models through APIs. These organisations frequently determine their fees based on token usage. The charges are usually calculated per million tokens processed.

DeepSeek itself initiated a significant price reduction in May. This action triggered a broader price war across the sector. The company had previously announced a permanent 75 per cent discount on its V4 API access. This substantial cut compelled rival companies, including ByteDance and Tencent Holdings, to introduce their own price decreases in response.

So this current price increment is slated to become effective when the official version of the V4 model is launched. The V4 model was initially released as a preview version in late Apr. The official launch is anticipated for mid July. This timeline was specified in the earlier mentioned email.

This reversal in pricing strategy marks a change from the trend observed among Chinese tech companies in recent months. Many have been reducing prices for their artificial intelligence models. The objective behind these previous reductions was to attract a larger base of developers and enterprise clients, given the intense domestic competition for AI market share.

And an AI price competition has been observed among various Chinese AI players. This competitive activity has been ongoing since earlier this month. The market has seen several Big Tech firms adjusting their pricing structures.

In mid June, TikTok's parent company, ByteDance, introduced a new offering. This was the Seedance 2.0 Mini video generation model. It was priced at 23 yuan for 1 million tokens. This cost represented approximately half of the standard version's price for the popular model.

But these actions from ByteDance followed earlier price cuts made by other prominent tech firms. Xiaomi had previously reduced its prices by up to 99 per cent on certain artificial intelligence models. Tencent Holdings also implemented significant reductions, slashing costs by as much as 70 per cent for some of its AI offerings.

• DeepSeek introduced peak hour surcharges for its V4 AI model API services.

• Peak hours are 9 a.m. to noon and 2 p.m. to 6 p.m. Beijing time.

• The V4 Pro peak hour rate is 12 yuan, or USD 1.77, per million output tokens, up from 6 yuan.

Source: SCMP