Maximilian228/Geode-Qwen3.6-35B-A3B

Geode · Qwen3.6-35B-A3B for phones

Model packs for Geode, an Android app that runs Qwen3.6-35B-A3B (35 billion parameters, 3 billion active per token) entirely on the phone: no cloud, no account. The experts are read from the phone's storage as they are needed, so the whole model does not have to fit in RAM.

Packs

Folder	Weights	Size	Quality (KL vs BF16 / top-1)	For
Q4/	every matrix 4-bit (GPTQ), embedding 8-bit, vision encoder 8-bit	20.8 GB (19.3 GiB)	0.096 / 90.9%	Snapdragon 8 Gen 2 / Gen 3 / 8 Elite, 12 GB+ RAM: the GPU (Adreno 7xx+) does the work, about 20 tokens/s

Quantization: GPTQ layer by layer (each layer on what the quantized layers before it produce), calibrated on chat, code, English and Ukrainian text, agent conversations with tool calls, the model's own reasoning and 30 languages (228k tokens). Quality is measured on held-out text of each kind (code, documents, English, other languages, an agent session, tool use, reasoning): KL divergence from the BF16 model and how often the top token is the same.

The packs are Geode's own format (geode.json describes the files; the experts in a GPU-friendly tiled layout). They are not GGUF and do not load in other runtimes.

Use

1. Install Geode (Google Play beta: geode.tifu.xyz).
2. Models → Download from Hugging Face → Maximilian228/Geode-Qwen3.6-35B-A3B → pick a pack.

Requirements: Android 11+, a 64-bit ARM processor with the dot-product instructions (ARMv8.2), and free storage for the pack.

License

Apache-2.0, as the base model: these packs are a quantized derivative of Qwen/Qwen3.6-35B-A3B by the Qwen team (see LICENSE). Answers come from an AI model and can be wrong.