ERNIE 5.1

Released 2026-05 · reasoning · 128K tokens · 0 benchmarks

Editorial notes

GA lanzada 9 mayo 2026 (anunciada 8 may, rollout en Create 2026 Beijing 13-14 may). Sub-red destilada de ERNIE 5.0 via "Multi-Dimensional Elastic Pre-training": ~1/3 params totales y ~1/2 activos vs ERNIE 5.0 (2.4T), a ~6% del costo de pre-training frontier comparable. Params exactos NO divulgados por Baidu. Proprietary/hosted-only (sin open weights). Pricing $0.59/$2.65 per MTok (Baidu Qianfan, max output 65536). Modalidad text (a diferencia del padre 5.0 multimodal — Baidu no confirma image/audio/video en 5.1 GA). Scores oficiales del blog sin numeros estandar comparables: LMArena Search Arena 1223 (#4 global, #1 China), AIME26 99.6 (con tools), supera a DeepSeek-V4-Pro en tau3-bench y SpreadsheetBench-Verified (sin valor numerico publicado). El score Text Arena 1476 pertenece al Preview (30-abr), NO a la GA.

Spec sheet

Company: Baidu
Country: CN
Type: reasoning
Release: 2026-05
Context: 128K tokens
Slug: ernie-5-1

No scores published for this model. The editorial notes have additional context.

Cite this model

BibTeX · APA

BibTeX

@misc{frontier-ernie-5-1,
  title  = {ERNIE 5.1},
  author = {{Baidu}},
  year   = {2026},
  note   = {Frontier Benchmarks AI atlas. Accessed 2026-06-17},
  url    = {https://frontierbenchmarks.com/models/ernie-5-1}
}

APA

Baidu (2026). ERNIE 5.1 [Large language model]. Frontier Benchmarks AI. Retrieved 2026-06-17, from https://frontierbenchmarks.com/models/ernie-5-1

Citation reflects the atlas page, not the original model paper. For the paper, see the "Resources" section above.

Battle vs another model ← All models More from Baidu Methodology