ERNIE 5.1
BaiduBaidu's fifth-generation flagship, released May 2026: a mixture-of-experts model with elastic depth, width and sparsity, aimed at text generation, reasoning, agentic tool use and creative writing, with a 128K context. Baidu positions it on cost efficiency, with roughly a third the total parameters and half the active parameters of the previous generation.
ERNIE 4.5 was released openly; the 5.x line is API-only and Baidu's Hugging Face organization lists no 5.x repository. Capability and adoption were re-sourced on 2026-08-14 to Baidu's quarterly results, which the release post does not carry. Verified 2026-08-13 via the ERNIE 5.1 release post and the Baidu organization page.
Openness
1 high confidence- weights
- closed
- data
- closed
- code
- closed
- license
- proprietary(Baidu Qianfan API)
No weights distributed for the 5.x flagship; closed API model.
- https://huggingface.co/baidu recorded 2026-08-13
Baidu HF org lists ERNIE-4.5 (21B-A3B-Thinking, VL-28B-A3B, VL-424B-A47B) and the ERNIE-Image models; no ERNIE 5.x repository, so no 5.x weights, corpus or code are distributed
Adoption
4 low confidenceBaidu's Q4/FY2025 earnings release states that ERNIE Assistant's MAU reached 202 million in December 2025 - a first-party, app-level figure for the ERNIE/Wenxiaoyan assistant rather than a 5.1-specific count, and the freshest published ERNIE user number: the Q1 2026 release gives no ERNIE MAU at all, only Baidu App at 655M for March 2026. The models are not on HF, so there is no download signal to band instead.
- https://www.prnewswire.com/news-releases/baidu-announces-fourth-quarter-and-fiscal-year-2025-results-302698026.html recorded 2026-08-14
Baidu Q4/FY2025 results, released 2026-02-26 - "ERNIE Assistant's MAU reached 202 million in December 2025"; for scale, "Baidu App's MAUs reached 679 million in December 2025"
- https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/ recorded 2026-08-13
ERNIE 5.1 release and API access. It carries no usage figure, so it does not support the MAU number the level rests on.
Capability
5 medium confidenceFrontier proprietary model. The arena standing is a vendor claim rather than a leaderboard read, which is why confidence stays medium: the 5.1 release post reports a score of 1,223, fourth place globally and first among Chinese models on the arena search leaderboard, and Baidu's quarterly earnings release states that ERNIE 5.1 ranked first among Chinese models on the text leaderboard and fourth globally on search. AIME 2026 with tool use at 99.6, and the GPQA and MMLU-Pro framing, are on the release post. A '#13 Text' placement that circulates elsewhere is supported by neither page, and a 128K context figure has evidence only on an ERNIE-4.5 card, a different model line, so neither is carried here.
- https://www.prnewswire.com/news-releases/baidu-announces-first-quarter-2026-results-302774476.html recorded 2026-08-14
Baidu Q1 2026 results, released 2026-05-18 - "Baidu launched ERNIE 5.1 in May 2026 ... Recently on LMArena, ERNIE 5.1 ranked first among Chinese models on the text leaderboard and topped the LMArena search leaderboard among Chinese models, ranking fourth globally." No ERNIE MAU figure on this release.
- https://ernie.baidu.com/blog/posts/ernie-5.1-0508-release/ recorded 2026-08-13
official MoE design and training-cost efficiency ("Pre-training Compute Cost at Only 6% of Comparable Models"); "On AIME26 (with tool use) ... ERNIE 5.1 scores 99.6 - second only to Gemini 3.1 Pro"; "on May 9, ERNIE 5.1 scored 1,223 to claim 4th place globally and 1st among Chinese models on the Arena Search leaderboard". No #13 Text ranking and no 128K context figure.
Verified 2026-08-13