Outlier-Ai/Outlier-70B-V3.3
Text Generation • Updated • 1.19k • 1
Six repos worth trying first: research flagship (70B V3.3), MLX shipping, Qwen3-Coder canonical, GGUF flagship, and V3.3 research.
Note Research flagship. 83.10% MMLU alpha-fixed. Ternary MoE overlay on Qwen2.5-32B.
Note Mac shipping flagship. 71.3 tok/s / 4.47 GB on M1 Ultra.
Note Top coding model for Apple Silicon. MLX 4-bit, 3B active.
Note Cross-platform GGUF flagship. llama.cpp / Ollama / LM Studio.
Note V3.3 on Qwen2.5-14B. 77.80% MMLU. Most-downloaded research.
Note V3.3 on Qwen2.5-7B. 70.87% MMLU. Entry research checkpoint.