Baidu has released ERNIE 5.5 Flash Base, a 280B parameter open-weight model, on Hugging Face. This model is a raw pre-training checkpoint and does not include chat capabilities out-of-the-box, requiring fine-tuning for conversational use. Running the model locally demands significant hardware, with estimates suggesting at least 144 GB of VRAM for a 4-bit quantization and up to 244 GB for a 1M context window. Baidu's license also restricts its use in the EU, UK, and South Korea, unlike the more permissive MIT license of DeepSeek V4 Flash Base, which ERNIE 5.5 Flash Base reportedly outperforms on many benchmarks. AI
IMPACT Requires substantial hardware for local deployment and has usage restrictions, potentially impacting adoption.
RANK_REASON Model release from a major AI lab (Baidu) with technical details and performance claims. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →