Baidu introduces ERNIE 5.5 Flash Preview with 280B MoE parameters and 1M context

Live via API on Baidu’s Qianfan platform with 14B active parameters per token and planned Apache 2.0 open weights, the post-trained text model targets long-horizon game code and procedural 3D CAD scripting.

Baidu released ERNIE 5.5 Flash Preview on 11 November 2026 as the first post-trained model built on ERNIE 5.5 Flash Base. Available immediately via API on Baidu AI Cloud’s Qianfan platform, the Mixture-of-Experts language model pairs 280 billion total parameters with 14 billion active parameters per token across a 1-million-token context window. Baidu stated that it plans to release the model weights under the Apache 2.0 licence at a later date, without naming a release date or publishing international USD token pricing.

Autonomous Unreal Engine 5.8 C++ and procedural Blender CAD builds

Trained specifically for multi-hour agent harnesses across game engines and parametric 3D software, ERNIE 5.5 Flash Preview completed an 11-hour autonomous coding run that merged multi-generation assets and wrote an entire playable third-person wuxia game (Yunxiu Xing) in C++ inside Unreal Engine 5.8. Because the model is text-only and cannot view rendered frames directly, it translated playtest glitches such as floating characters and camera offsets into automated code assertions, and separately generated nearly 100 playable 2D and 3D mobile mini-games from design documents using deterministic state-space verification.

In 3D modelling demonstrations, coordinated ERNIE 5.5 Flash Preview agents wrote Python scripts in Blender 5.2 (bpy) and FreeCAD to assemble a parametric airport and dual-deck airliner rendered with Cycles and OptiX, reconstruct the Horyu-ji temple complex, block out a 2,100-object, 1.82-million-triangle previsualisation scene, and model articulated W16 engine and lunar rover assemblies.

Technical details

Model architecture

280B total / 14B active Mixture-of-Experts text language model with a 1M-token context window

Access and weight plan

Live on Baidu AI Cloud Qianfan API; Apache 2.0 open weights planned (date and USD pricing undisclosed)

Autonomous game showcases

11-hour Unreal Engine 5.8 C++ game (Yunxiu Xing) plus nearly 100 playable 2D and 3D mobile mini-games

Procedural 3D CAD targets

Blender 5.2 (bpy with Cycles and OptiX) and FreeCAD up to 2,100-object, 1.82-million-triangle scenes

Release details & context

Blind text-to-assertion verification for 3D and game engine builds

Announced on Baidu’s ERNIE 5.5 Flash Preview release page, the first post-trained checkpoint from ERNIE 5.5 Flash Base takes a distinct architectural route from Baidu’s multimodal Unlimited OCR 2.0 and visual spatial studios like Moonshot AI’s Kimi 3D Make.

Because the 14B-active MoE model accepts text alone and never views rendered screenshots, its multi-agent harnesses convert spatial defects such as floating characters, camera offsets and mesh collisions into deterministic state-space and C++ or Python assertions across Unreal Engine 5.8 (complementing Epic’s agent-scriptable Unreal Engine 5.9 MegaJam Preview), Blender 5.2 and FreeCAD assemblies including Horyu-ji, a dual-deck airliner, a lunar rover and an articulated W16 engine.

Common questions

How does text-only ERNIE 5.5 Flash Preview debug visual 3D and Unreal Engine glitches?

Rather than inspecting rendered viewport frames, its agent harnesses encode spatial defects such as floating characters and camera offsets as deterministic C++, Python and state-space code assertions.

Did Baidu release ERNIE 5.5 Flash Preview open weights on 11 November 2026?

Not on release day: ERNIE 5.5 Flash Preview launched via API on Baidu AI Cloud’s Qianfan platform, with Apache 2.0 open weights promised for a later unspecified date.