YOON1v/Apex-3
just landed · new, unvettedmodel
New, unvetted: created or first seen in the last few days, not from a trusted org and not past the signal threshold yet.
Apex-3 is a mixture-of-experts language model with 21B total and 3.36B active parameters, trained from scratch with no external pretrained weights. Mixture-of-experts, 21B total, 3.4B active (64 experts, 4 per token). Context: 8K tokens (config). License: apache-2.0.
- Org
- YOON1v
- Released
- 2026-10-28
- Status
- just landed · new, unvetted
- License
- apache-2.0 (permissive)
- Parameters
- 21.0B total, 3.4B active
- Architecture
- MoE · 64 experts, 4/token · Apex3ForCausalLM
- Context
- 8,192 tokens
- Task
- text-generation · text
- Signal
- 3 HF likes · 0 downloads
Runs on
mixed: real files + estimate; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~15.4 GB (estimated size): 24GB GPU (3090 / 4090), 24GB GPU (7900 XTX), Mac 24GB
- Q8 ~25.1 GB (estimated size): 32GB GPU (5090), Arc B70 32GB, Mac 36GB, Mac 48GB, CPU only, 32GB RAM (slow)
- FP16 ~44.8 GB: 48GB GPU (RTX 6000 Ada / A6000), 96GB GPU (RTX Pro 6000 Blackwell), 1× DGX Spark (128GB), 2× DGX Spark (256GB), Strix Halo 128GB unified (~96GB usable), Mac 64GB (M1 Max), Mac 96GB, Mac 128GB, Mac 192GB, Mac 256GB, Mac 512GB, CPU only, 64GB RAM (slow), CPU only, 128GB RAM (slow), CPU only, 256GB RAM (slow)
Links
First published here 2026-10-28 · updated 2026-10-28 · markdown · JSON