Baidu launches ERNIE X1, its first reasoning model

Background

A reasoning model is an AI model that works through a problem in several internal steps before it gives an answer. Developers often call this deep thinking or chain of thought. The extra steps usually make such a model slower and more expensive to run than a plain chat model, but they tend to help with questions that need multi-step logic, such as mathematics or programming. Baidu's ERNIE family of large language models (LLMs) had already been released in several versions, and ERNIE Bot (文心一言) was the chatbot through which ordinary users reached them.

What happened

Baidu launched ERNIE X1 on 16 March 2025 as a deep-thinking reasoning model. The company called it its first multimodal reasoning model capable of tool use.^[1] It was unveiled in the same announcement as ERNIE 4.5.^[1]

Baidu claimed that X1 matched DeepSeek-R1 at half the price, quoting RMB 0.002 per thousand input tokens and RMB 0.008 per thousand output tokens.^[1] A token is the unit of text a model reads and writes, so the price per thousand tokens sets what a developer pays for each request. VentureBeat converted the same prices to $0.28 per million input tokens and $1.10 per million output tokens, compared with $0.55 and $2.19 for DeepSeek-R1.^[2]

Key details

ERNIE X1 supported advanced search, document question answering, image understanding, AI image generation, a code interpreter and webpage reading.^[1] Baidu did not publish specific benchmark results for X1 at launch.^[3] X1 was free to use on ERNIE Bot, and API access through Qianfan was to follow.^[1]

Impact

Coverage of the launch concentrated on cost. VentureBeat's headline described the models as undercutting DeepSeek and OpenAI on price.^[2] Built In placed the release in the context of the US-China AI race.^[3] X1 became the starting point of Baidu's reasoning-model line, and later versions, including X1 Turbo and X1.1, were compared against it.