# Mlx Bench

> Measure a locally served MLX model without downloading it.

- Skill: `cavi-ai/mlx-bench-4` (Agent Skill, multi-file: 45 files)
- Install (CLI): `npx get-skill add cavi-ai/mlx-bench-4`
- Raw SKILL.md: https://api.skillmd.com/api/skills/cavi-ai/mlx-bench-4/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: cavi-ai (https://skillmd.com/u/cavi-ai)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/cavi-ai/mlx-bench-4

---


# MLX Bench

Resolve `<skill-dir>` as the absolute directory containing this SKILL.md. Never resolve the bundled executable from the shell working directory.

canonical capability ID: mlx-agent.bench

Measure a model already served by a running local runtime:

`python3 <skill-dir>/scripts/mlx-agent bench run --repo <repo> --runtime <runtime> --json`

Present the returned measurements as returned. Bench must not start servers, download model weights, or change configuration; the model must already be served by an existing local runtime.

