169Pi's Alpie Seed is a 9-billion-parameter multimodal model built by post-training Qwen3.5-9B, with support described for text, images and video. The publisher's model card lists Apache 2.0 and a 262,144-token context window, alongside full-precision and separately distributed quantized formats.
This is an adaptation of an existing Qwen model, not a newly developed base architecture. 169Pi says it used supervised fine-tuning and reinforcement learning in its post-training. Its technical report documents the design and evaluations.
Small weights do not settle the whole hardware question
The model card lists multiple GGUF builds and Ollama variants. It identifies a 5.6 GB text-oriented default and a 6.6 GB vision variant; those are package sizes, not complete system memory requirements. Longer context, image processing, runtime choices and other applications can increase memory use substantially.
Developers interested in image understanding should check the exact variant: the card lists text input for some Ollama tags and text plus image for its vision tag. Available input modalities can differ between packages and software implementations.
Read the benchmarks carefully
169Pi reports benchmark results for its FP16 model, but says the comparisons draw on figures published elsewhere rather than a single consistent rerun. Prompts, evaluation harnesses and versions differ. Those numbers therefore deserve closer inspection before treating them as a ranking, and they should not be assumed to apply to lower-bit versions.
The practical appeal is an accessible model family with a clearly stated lineage and multiple deployment formats. Whether its particular tradeoffs are useful depends on the task and hardware. AiBrainWorX has not run it; our model profile links to the original files and terms.
Original reporting; primary sources reviewed November 14, 2026. No independent hands-on test claimed.
