Simple pricing.

Compaxis SQ4 uses a more efficient compression algorithm than the standard Q4_K_M. The result is a 4-bit GGUF that's measurably closer to FP16 — 21% to 78% closer across our 7-model production line — at roughly the same file size, same speed, and same loader. Drop it into llama.cpp / Ollama / LM Studio just like any other GGUF.

Real GGUF. Compaxis SQ4 files load in unmodified llama.cpp, Ollama, LM Studio, KoboldCpp, and anything else that already reads GGUF. No new runtime to install, no API to integrate. The Hub is a convenience download manager — you can also drop the .gguf into your existing tools by hand.

Same as any 4-bit GGUF: roughly the model file size + a couple of gigabytes of context cache. Qwen 3 1.7B SQ4 runs comfortably on a 4 GB GPU. Qwen 3 8B SQ4 fits in 8 GB. Qwen 3 14B SQ4 wants 12 GB. The 27B-class models want 20–24 GB. Compaxis SQ4 files are about 15% larger than Q4_K_M on average — the bits we add buy the quality.

Yes. Your Pro or Lifetime license covers commercial use of the SQ4 files we publish. Note that each underlying model has its own upstream licence (Apache 2.0 for Qwen and Gemma 4, MIT for the DeepSeek distill) — Compaxis re-quantizes them but doesn't change those upstream terms. We list each model's licence on its download page.

Any Compaxis SQ4 GGUF you downloaded while your subscription was active is yours to keep and use. The Hub stops pulling new updates and stops downloading additional Pro models, but every file already on disk keeps working in llama.cpp.

Lifetime covers every SQ4 model we release during the V1.x series (V1.0, V1.1, V1.2 …). If we ship a V2.0 with a fundamentally different product — for example moving beyond GGUF, or new quantization formats — there may be a discounted upgrade price. The V1.x line itself is included indefinitely.

Not yet for V1. If you need 5+ seats or a site licence, email and we'll work something out.