Tencent Releases ‘Turbo’ AI Model To Counter DeepSeek

Chinese tech giant Tencent has released a new AI model that it said can answer questions faster than DeepSeek’s R1, in the latest of a surge of new model launches since DeepSeek’s popularity caused a global rethink of the AI race.

Tencent’s Hunyuan Turbo S is designed specifically for high-speed responses, in contrast to DeepSeek’s R1, Tencent’s own Hunyuan T1, and “other slow thinking models that need to ‘think for a while before answering'”, Tencent said in a statement.

Reasoning models such as R1 or OpenAI’s o3-mini are designed to take more time before delivering responses, using a technique intended to improve accuracy.

Price cut

Tencent claimed that Turbo S’ capabilities matched DeepSeek’s V3 large language model in areas such as knowledge, math, and reasoning.

The company also said the usage costs for Turbo S were many times cheaper than previous iterations, illustrating how DeepSeek’s low-cost, open-source approach has forced competitors to lower prices.

Chinese tech giants have rushed to roll out new AI models in the past few weeks, with Alibaba releasing the Qwen 2.5-Max model that it said outperforms DeepSeek-V3 in all categories.

Alibaba also said it would spend massively on AI development over the next three years.

Chinese search giant Baidu said last month it would shift its Ernie LLM to an open-source development model from 30 June, in a major shift of strategy.

Open source shift

The company’s founder and chief executive Robin Li had long argued closed-source is the only viable model for AI development.

Baidu also said last week the Ernie Bot service would be free to use from 1 April, ending a 17-month trial at charging a fee for the service.

DeepSeek’s app rose to international prominence in late January, roiling international stock markets as investors were forced to rethink the logic behind leading tech companies’ massive AI spending.

DeepSeek itself last week released technical details of its high-efficiency AI training methods in a series of open-source projects.