LIVE
CHINA AI

Baidu launches 280B-parameter ERNIE 5.5 Flash MoE model and ERNIE Evolve automated AI research engine

Trained on 33.5 trillion tokens with 14B active parameters per token, Baidu’s new sparse Mixture-of-Experts model delivers 200+ tokens per second across a 1-million-token context window.

UpdateNext Staff
Editorial team
4 min read
1 view
Baidu ERNIE 5.5 Flash Base and Preview launch
Baidu released both the raw ERNIE 5.5 Flash Base pre-training checkpoint for developers and the high-speed ERNIE 5.5 Flash Preview on OpenRouter. (Image: Unite.AI / Baidu)

Baidu has escalated China’s price-performance AI war with the release of ERNIE 5.5 Flash Base and ERNIE 5.5 Flash Preview—opening up both the raw pre-training base checkpoint for custom enterprise post-training and a ultra-fast hosted inference endpoint on OpenRouter and Baidu Qianfan.

Built on a highly sparse Mixture-of-Experts (MoE) architecture, ERNIE 5.5 Flash packs 280 billion total parameters while activating just 14 billion parameters on each forward pass. Pre-trained across a 33.5-trillion-token multilingual and code corpus, the model supports a full 1-million-token context window and sustains generation speeds above 200 tokens per second.

ERNIE Evolve: Autonomous algorithm discovery and R&D loops

Alongside the ERNIE 5.5 Flash weights, Baidu Research unveiled ERNIE Evolve, an evolutionary coding and research agent that combines large language models with automated verifier loops to discover faster sorting kernels, optimize logistics routing, and run iterative machine-learning experiments without human intervention.

✓ WE KNOW
  • Baidu has released ERNIE 5.5 Flash Base and ERNIE 5.5 Flash Preview, featuring 280 billion total parameters with only 14 billion active parameters per token.
  • Pre-trained on 33.5 trillion tokens, the model supports a 1-million-token context window while generating over 200 tokens per second.
  • Baidu also introduced ERNIE Evolve, an automated scientific and engineering research system that formulates hypotheses, writes code, and evaluates experimental outcomes autonomously.
? STILL UNCLEAR
    ADVERTISEMENT
    Get the daily AI briefingSubscribe
    “By activating only 5 percent of its 280B parameter pool per token, ERNIE 5.5 Flash delivers frontier-grade reasoning at a fraction of the inference cost required by dense models.”
    — Baidu AI Cloud Engineering Note, November 2026
    SOURCES:Unite.AI — Baidu releases ERNIE 5.5 Flash Base for further training ↗CryptoBriefing — Baidu launches ERNIE 5.5 Flash Preview with 280B parameters ↗
    > how_we_reported_this
    • ✓Sources: written from primary sources and official statements, linked above.
    • ✓Fact-check: names, dates and claims checked against the original sources.
    • ✓AI use: researched with AI tools; written and reviewed by our editors. Our AI policy
    • ✓Independence: UpdateNext has no financial relationship with the companies covered.
    1 view
    > react:
    UpdateNext Staff
    The UpdateNext newsroom covers AI and tech from primary sources, every day. Every story is checked against official statements.
    Write for us →
    Comments
    Comments are moderated · Sign in to comment
    > read_next
    ENTERPRISE AI SHOWDOWN
    Alibaba launches Qwen Office 2.0 with autonomous 'Team Agents' across DingTalk as China’s AI office user base tops 140 million
    CLOUD TOKEN WARS
    China’s Model-as-a-Service (MaaS) market surges past 106 billion yuan as ByteDance Volcano Engine and Alibaba Cloud battle on tokens
    HUAWEI FULL-STACK AI
    Huawei debuts Xiaoyi Work AI agent on HarmonyOS PC and scales Ascend SuperNode clusters for DeepSeek V4-Flash workloads
    What's next in AI.

    The daily briefing on AI news, tools and breakthroughs.