ModelDen

MiniCPM-SALA

9.5B · Q4_K_M · ~5.8 GB
Released February 11, 2026

How good is it? — per-capability quality

Each number is the model’s score on its best public benchmarkfor that skill — named and linked below, never a blend. Scores compare cleanly when the benchmark matches; benchmarks are gameable, so we show the receipts rather than hide behind one number.

Coding61
LiveCodeBench v5 · source · as of 2026-06
Math & logic78
AIME 2025 · source · as of 2026-06
General67
MMLU-Pro · source · as of 2026-06

MiniCPM-SALA (Sparse Attention and Linear Attention) is a 9B hybrid model built from a MiniCPM-4.…

About this model

Context window
32,768 tokens
Will MiniCPM-SALA actually run on YOUR graphics card?

This page shows fit for a generic VRAM tier. Tell us your GPU for a personalized speed estimate and fit verdict.

Enter your graphics card →