Baidu debuts its first AI reasoning model to compete with DeepSeek – News

The Chinese search engine giant Baidu Inc. is stepping up its game in artificial intelligence with the launch of two of its most powerful models yet, including a new reasoning-focused model that it claims can outperform DeepSeek Ltd.’s R1.

Baidu first debuted its Ernie series of models in 2023, before unveiling a “significant upgrade” the following year with the launch of Ernie 4.0 Turbo.

Now, it’s lifting the lid on Ernie X1, which it says delivers performance on par with DeepSeek R1 at just half the price of its rival. According to Reuters, it boasts “stronger understanding, planning, reflection and evolution capabilities,” and it’s also the first “deep-thinking” model that’s able to use independent tools autonomously.

DeepSeek has emerged as the new standard for AI model makers to beat after launching R1 just before Christmas. As a reasoning model, R1 stood out for matching the capabilities of the best U.S.-made AI models, despite its maker only spending a small fraction of the cost its rivals had spent in training it.

As for Baidu’s other model, Ernie 4.5, it’s a more standardized, nonreasoning model that’s designed to compete with OpenAI’s GPT-4o, yet it boasts “excellent multimodal understanding” and has a “more advanced language ability,” the company said. Its ability to understand humans, generate responses and its memory have also been improved.

Multimodal AI systems are notable for their ability to process various different types of data. They can, for example, ingest text-based prompts, videos, images or audio files, and also convert content across these formats.

In addition to being multimodal, Ernie 4.5 also reportedly has a “high EQ,” which helps it better understand internet memes and satirical cartoons, Baidu claimed.

The emergence of DeepSeek re-energized the AI race and Baidu in particular will be optimistic that it can improve its fortunes. The company was one of the first Chinese firms to launch a ChatGPT-style AI model, but it has struggled to compete even in its domestic market, despite claiming that its earlier Ernie models are on par with OpenAI’s best.

The Beijing-headquartered company faces intense competition from rival tech firms at home, including the TikTok parent company ByteDance Ltd. and Tencent Corp., whose rival chatbots have gained many more users.

However, all of China’s leading AI firms were upended by the dramatic debut of DeepSeek, which took the industry by storm with R1, paving the way for an era of much more affordable AI models. Since then, China’s government and dozens of local companies have backed DeepSeek, integrating its model within their own products and services, leaving its rivals to play catch-up. Even Baidu has adopted DeepSeek R1, iterating the reasoning model with its iconic search engine.

DeepSeek’s rivals are determined to step up to the plate, and in February Tencent’s WeChat debuted a new model that was claimed to be able to respond to queries even faster than DeepSeek.

Also last month, Alibaba Group Holding Ltd. said it would be investing 380 billion yuan (around $52 billion) into its AI and cloud computing research efforts over the next three years.

Meanwhile, Baidu has followed DeepSeek’s lead, saying it intends to open-source the codebase of its Ernie models later this year.

Photo: Baidu