Baidu Releases ERNIE 5.5 Flash Preview with 280 Billion Total Parameters and Support for a 1 Million Token Context
Baidu has released the foundation model ERNIE 5.5 Flash Preview, with 280 billion total parameters, 14 billion parameters activated per token, support for a context of up to 1 million tokens, and a generation speed exceeding 200 tokens per second. The model has undergone specialized training for scenarios such as programming, 3D modeling, knowledge work, and content creation, and has been launched on Baidu AI Cloud's Qianfan platform for developers to call via API. API pricing is 2 yuan per million input tokens during peak hours and 1 yuan during off-peak hours, and 8 yuan per million output tokens during peak hours and 4 yuan during off-peak hours, with a limited-time 50% discount. The model weights will be released to the global open-source community on November 25.
Why read this
SelectedAI score 77
The original text provides the parameter scale, context length, and API pricing, allowing readers to assess the model's position in long-context and low-cost scenarios.
Topics
Body · AI translation
On November 11, Baidu's latest foundation model, ERNIE 5.5 Flash Preview, was officially released. ERNIE 5.5 Flash Preview has a total parameter count of 280 billion, activates 14 billion parameters per token, supports a context length of up to 1 million tokens, and performs outstandingly in areas such as programming, 3D modeling, knowledge work, and content creation. Currently, in multiple benchmark tests including Terminal-Bench and AutomationBench, ERNIE 5.5 Flash Preview has achieved competitive scores while maintaining extremely fast decoding speed.
Programming capability is a major highlight of the new ERNIE model. ERNIE 5.5 Flash Preview covers a wide range of coding scenarios, from fixing and iterating in real code repositories to fully building complex applications. Especially in long-cycle development tasks, the model demonstrates stronger planning, execution, and continuous iteration capabilities, which has also greatly improved the game development performance of ERNIE 5.5 Flash Preview. The model has undergone specialized training for game development scenarios and is adapted to mainstream development tools, completing game design, feature production, and scene construction according to users' ideas and needs, making it easier for users to turn game ideas into playable works.
At the same time, the new ERNIE model is particularly outstanding in parametric modeling and spatial reasoning. By strengthening specialized training in the use of tools and software such as Blender and CAD, it has significantly improved the 3D modeling and design capabilities of ERNIE 5.5 Flash Preview. With tool assistance, ERNIE 5.5 Flash Preview can accurately understand and infer the shape, proportions, and spatial relationships of objects from text descriptions, and transform them into precise 3D models through procedural modeling. At the same time, it can flexibly integrate and reuse existing assets, uniformly integrating external materials into existing workflows to create 3D scenes with grand scale and rich details.
This capability also extends to the fields of professional knowledge work and content creation. ERNIE 5.5 Flash Preview can complete multi-step work according to user needs and deliver highly professional spreadsheets, documents, or presentations. For fields such as finance and law, it has built 300,000 workspaces and 10 million reference files, covering more than 40 file types, with training environments ranging from personal workspaces of 10 to 100 files to enterprise knowledge bases composed of thousands of files.
For content creators' video creation needs, ERNIE 5.5 Flash Preview can collect, synthesize, and arrange materials, adjust visuals and pacing, and output finished videos. For scenarios such as popular science, comic drama, creative editing, and hot-topic interpretation, it has built more than 150,000 creation environments, covering the complete process from idea generation, scripts, storyboards, visual materials, and voiceover, adapted to thousands of creation skills, with a cumulative total of about 5,300 minutes of finished videos.
It is understood that at the underlying technology level, ERNIE 5.5 Flash Preview, through multiple technological innovations such as a new hybrid sparse attention architecture, multi-router MoE, and ECHO multi-token prediction, improves model capability while reducing computation and memory overhead, supports million-token long context, and achieves a generation speed of more than 200 tokens per second. In addition, ERNIE 5.5 Flash Preview introduces self-developed RSI technology for the first time, with ERNIE Evolve participating in model inference optimization, exploring a new path for AI to participate in model research and optimization.
Currently, ERNIE 5.5 Flash Preview has been launched on Baidu AI Cloud's Qianfan platform, and developers can call the model via API. API pricing is 2 yuan per million input tokens during busy hours and 1 yuan during idle hours, and 8 yuan per million output tokens during busy hours and 4 yuan during idle hours, with a limited-time 50% discount, giving it strong price competitiveness among models of the same parameter scale. In addition, the model weights will be officially released to the global open-source community on November 25.