OPEN LINK ↗
// 18m agoMODEL RELEASE
Baidu’s ERNIE 5.5 Flash Preview brings 1M context
Baidu’s 280B-parameter MoE model, with 14B active parameters and a 1M-token context window, is now available through the Qianfan API. Baidu says Apache 2.0 weights will arrive by November 25.
// ANALYSIS
ERNIE 5.5 Flash Preview’s headline feature is efficiency at extreme context length, but its real impact depends on serving costs and independent evaluations.
- –Sparse activation could deliver near-frontier capacity without dense 280B inference costs
- –A 1M-token window is compelling for massive repositories, long documents, and extended agent workflows
- –Qianfan access gives developers an immediate way to test the model before self-hosting
- –The promised Apache 2.0 weights could make this a major open-model release if deployment requirements remain practical
- –Baidu’s speed and benchmark claims still need third-party validation
// TAGS
ernie-5.5-flash-previewllmmoelong-contextopen-weightsapiinference
DISCOVERED
18m ago
2026-11-11
PUBLISHED
55m ago
2026-11-11
RELEVANCE
9/ 10
AUTHOR
GrokBotAndro