百度发布文心大模型 4.0
国产旗舰对标 GPT-4 的正面冲击
百度在 2023 百度世界大会上发布文心大模型 4.0,宣称综合能力与 GPT-4 相当,并接入搜索、地图、网盘等全线产品。它被视为国产大模型对标国际旗舰的标志性一役。
2023 年 10 月,百度世界大会上,李彦宏宣布发布文心大模型 4.0,并说了一句分量很重的话:「综合水平与 GPT-4 相比毫不逊色。」在 2023 年那个国产大模型普遍被视为「追赶者」的年代,这是国产旗舰第一次公开、正面地宣称对标当时全球最强的模型。
半年前,文心一言作为「中国版 ChatGPT」率先登场,抢占了先发声量,但也背负着「不够强」的质疑。接下来的半年,百度一边迭代技术,一边把文心接入搜索、地图、文库、网盘等自家产品。到 4.0 发布时,李彦宏现场演示了复杂推理与多轮对话,试图用实际表现回应「国产模型行不行」的拷问。
文心 4.0 的定位是「全面的能力提升」:理解、生成、逻辑、记忆四个维度一起升级。更重要的是,它被定位成百度的「AI 底座」——不是发布一个模型就完了,而是要把能力铺进百度全线产品。搜索开始给答案、地图开始懂对话、文库开始能创作,一家公司的 AI 化改造,以文心为中枢全面展开。
从行业视角看,文心 4.0 的意义在于它正式开启了国产旗舰「对标 GPT-4」的竞赛。此后一年里,各家国产大模型——通义千问、GLM、DeepSeek——都以「对标国际前沿」为标尺展开竞争。这种「正面叫板」的姿态,不管最终是否兑现,本身就改变了国产 AI 的叙事:从「求同存异」到「正面竞争」。
回看文心 4.0 的发布,它的价值在于一次态度宣示:中国大模型不再满足于「可用」,而是敢于公开对标全球最强。今天国产模型已经整体站上世界舞台,而回看这个过程的起点,文心 4.0 那句「毫不逊色」的宣言,正是第一声号角。
In October 2023, at Baidu World, Robin Li announced ERNIE 4.0 with a weighty claim: "comprehensive capability on par with GPT-4". In an era when domestic models were widely seen as chasers, this was the first time a Chinese flagship publicly and directly claimed to match the world's strongest model at that moment.
Six months earlier, ERNIE Bot had debuted as "China's ChatGPT", grabbing first-mover attention while carrying doubts about its strength. For the next six months Baidu iterated on technology while wiring ERNIE into search, maps, Wenku, and cloud storage. By the 4.0 launch, Li demoed complex reasoning and multi-turn dialogue on stage, trying to answer the question "are Chinese models good enough" with actual performance.
ERNIE 4.0's positioning was comprehensive capability gains across understanding, generation, logic, and memory. More importantly, it was positioned as Baidu's AI base—not just releasing a model, but spreading its capability across the entire product line. Search started giving answers, Maps started understanding dialogue, Wenku started creating. A company's AI transformation unfolded with ERNIE at the center.
From an industry view, ERNIE 4.0 formally started the domestic "match GPT-4" race. In the following year, domestic flagships—Qwen, GLM, DeepSeek—all competed against the "international frontier" yardstick. That posture of head-on challenge, whether or not fully delivered, changed the domestic narrative from "find our own way" to "compete directly".
Looking back at ERNIE 4.0's launch, its value was a statement of attitude: Chinese large models would no longer settle for "usable" but dared to publicly challenge the global best. Today domestic models stand on the world stage as a whole, and at the origin of that climb is ERNIE 4.0's declaration of being "on par"—the first trumpet call.
展开完整事件档案人物、主题、模型与产品
- 人物
- —
- 模型
- —
- 产品
- —