Is ERNIE 5.5 Flash Preview state of the art?Pending benchmarks.Announced Nov 11, 2026; no independent results yet.
Baidu's post-trained ERNIE 5.5 Flash checkpoint, built on the ERNIE 5.5 Flash Base released November 5, 2026. A sparse MoE language model with 280B total parameters and 14B active, and a 1M-token context. Baidu describes it as an intermediate post-training checkpoint — training is not finished — aimed at coding, 3D modeling, knowledge work and content creation. The API is live on Qianfan (model ernie-5.5-flash-preview). Weights are not published yet; Baidu says they will be open-sourced under Apache 2.0 soon, and by November 25, 2026 on the official Chinese release page.
The details so far
- Lab
- Baidu
- Announced
- Nov 11, 2026
- API
- Live · ernie-5.5-flash-preview
- Open weights
- Announced, not yet published · Not released yet. No public Hugging Face Preview repo found. The English release page says weights will be open-sourced under Apache 2.0 soon; the official Chinese page says by November 25, 2026
- Parameters
- 280B total / 14B active
- Type
- Post-trained sparse MoE language model on ERNIE 5.5 Flash Base; Baidu calls it an intermediate checkpoint because post-training is still ongoing (Baidu release page)
- Context
- 1M-token context (Baidu release page and Qianfan model list)
- Hosted access
- Qianfan API, model ernie-5.5-flash-preview (Baidu release page and Qianfan model list). The official release page also links OpenRouter
- Pricing
- Official Qianfan China billing: promotional peak (08:00–22:00) 0.001 yuan input / 0.004 yuan output per thousand tokens, off-peak half that, cache hit one-tenth of the window's input rate; standard list prices are twice the promotional rates. Official international Qianfan price list: $0.28 input / $1.13 output / $0.028 cache hit per million tokens
- License
- Apache 2.0 announced for the forthcoming weights (English release page). Not yet applicable to a public checkpoint
- Related
- Separate from ERNIE-5.5-Flash-Base (the November 5 pre-training checkpoint) and from earlier Ernie 5.1 / 5.0 API snapshots
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: self-reported as an intermediate post-training checkpoint that is already competitive with models of its size, but because training is not finished, some capabilities may still fall short (Baidu release page).
Sources
LABBaidu ERNIE: Introducing ERNIE 5.5 Flash Previewernie.baidu.com/release/ernie-5-5-flash-preview/ · Nov 11, 2026LABBaidu ERNIE: ERNIE 5.5 Flash Preview (Chinese)ernie.baidu.com/zh/release/ernie-5-5-flash-preview/ · Nov 11, 2026LABQianfan model service billingcloud.baidu.com/doc/qianfan/s/wmh4sv6yaLABQianfan international price listintl.cloud.baidu.com/en/doc/qianfan/s/Jm8r1826a-intl-enLABQianfan international model listintl.cloud.baidu.com/en/doc/qianfan/s/7m95lyy43-intl-enTracking continues until an independent board lists it.