According to the model list updated by Baidu Intelligent Cloud Qianfan AI Application Developer Center on November 12, 2025, Wenxin 5.0 (ERNIE 5.0) has appeared in the recommendation area as a "flagship model" and is clearly described as a "Wenxin new generation model" and a "native full-modal large model". The document states that Wenxin 5.0 adopts native omni-modal unified modeling technology to jointly model text, images, audio, and video, and has comprehensive omni-modal capabilities, so the expression "natively omni-modal foundational model" is basically consistent with the official technical positioning.
The capability description of Wenxin 5.0 on the same page includes: the basic capabilities have been fully upgraded, the overall performance in benchmarks has been improved compared to the previous generation, and the multimodal understanding, instruction following, creative writing, factuality, agent planning and tool application are "particularly good", and the context length has been extended to 64K or 128K to better support long document understanding and complex reasoning. This is consistent with the connotation of slogans such as "good at full-modal understanding, creative writing and instruction following". However, the current access point names that can be called on the Qianfan platform are "ERNIE-5.0-Thinking-Preview" and "ernie-5.0-thinking-latest", and the official ERNIE blog is still based on "ERNIE-5.0-Preview-1022", emphasizing that it is being tested on LMArena and plans to "officially release it in the near future", indicating that it is still in the preview and phased opening stage. rather than completely universal.
As for the commitment to "continue to invest and develop more cutting-edge models to promote the boundaries of intelligence", Baidu has repeatedly reiterated in recent years in financial reports, technical white papers and many press conferences that it will increase R&D investment in large models, Kunlun core chips and AI infrastructure for a long time, and external reports also regard it as a strategic focus for the company's future growth. Judging from public information, this kind of statement is a continuous corporate strategy statement, not an isolated new commitment. Based on multiple sources, it can be determined that ERNIE 5.0 is positioned as a new generation of native full-modal basic model of Wenxin, and its self-evaluation of multi-modal understanding, creative writing, instruction following, and other capabilities is consistent with the official technical documentation, but it is still in the "Preview/phased opening" stage, and the specific scope and timetable of full opening are still to be officially announced later.
FAQs
Q: Has ERNIE 5.0 really been released?
A: Baidu Qianfan documentation has listed Wenxin 5.0 as the flagship model and provides the "ERNIE-5.0-Thinking-Preview / -latest" access point, while the ERNIE official blog mentions that "ERNIE-5.0-Preview-1022" is tested on LMArena and will be "officially released in the near future". It can be considered that the model has entered the preview and open testing stage, but whether it is fully open to all users and the specific commercial strategy is not fully clear.
Q: Is the phrase "foundational model" reliable?
A: Qianfan's official page clearly states that "Wenxin 5.0 is a native full-modal large model", and explains that the use of unified modeling technology to jointly model text, images, audio, and video is positioned as a basic general model rather than only for a certain vertical scene, so the statement of "natively omni-modal foundational model" is basically consistent with the official positioning.
Q: Does it really "outstand" in multimodal understanding, creative writing, and instruction following?
A: The original Baidu document states that Wenxin 5.0 performs "particularly well" in multimodal understanding, instruction compliance, creative writing, factuality, and agent planning and tool application, and emphasizes that the benchmark results are better than the previous generation. These are still self-reported results by manufacturers, and external independent evaluations are currently mainly from platforms such as LMArena rankings, which are still in the process of continuous verification.
Q: How is ERNIE 5.0 now available to general developers?
A: Currently, API access is provided in the form of "ERNIE-5.0-Thinking-Preview/Latest" through the Baidu Intelligent Cloud Qianfan platform, which is aimed at enterprises and developers; There may be differences between accounts, regions, and billing policies, and we have not yet seen a unified policy description that is "completely free and fully open to all end users".
Q: Is Baidu's "will continue to invest in cutting-edge models" empty words or is it really based?
A: Baidu continues to disclose its high-intensity investment in AI research and development, including large models, AI chips, computing power platforms and industry solutions, in financial reports, patent white papers and many conferences, and the cumulative AI-related R&D expenditure over the years has reached hundreds of billions of yuan in public data. Combined with the statements of the latest World Conference and the Cloud Intelligence Conference, "continuing to increase investment and expand the boundaries of intelligence" belongs to its long-term strategic main line, not a one-time publicity slogan.
