The public inclusionAI/Ling-3.1-flash repository history clarifies the development: October 13 commits add files, and the corresponding tree contains safetensors weight shards. This is more specific evidence than repository creation alone. It documents a weight upload this week without establishing the model’s first publication across every platform.
The card describes a hybrid mixture-of-experts reasoning model with a declared 560 billion parameters, 25 billion active per token and a one-million-token context window. Those are publisher specifications, not independent measurements. The historical October 14 card carries an MIT licence tag; a standalone LICENSE file was not found at the checked address. Reuse terms therefore warrant review rather than treating this check as legal clearance.
OpenRouter dates its own offering to October 2. Earlier hosted access is compatible with a later weight upload: an API and downloadable files are different events. Current SGLang and vLLM recipes target substantial multi-GPU configurations; Flash does not mean the complete model fits on an ordinary computer. inclusionAI’s benchmark and compatibility claims still need evaluation in the selected environment.