bench.pt

bench.pt is a command for coding agents from druide67/asiai. It costs 29 tokens per session (1,767 once invoked), scanned B, original, Apache-2.0.

A command-line benchmark for testing AI models side by side across local engines on Apple Silicon Macs. It uses standard prompts to measure output speed, first-response time, memory use, temperature, and power efficiency.

In plain words
What is it for?
Use it to test coding, tool-use, reasoning, or long text generation; compare models and engines; check power use; export JSON; view history; or share results with a community database.
Why use it?
It removes guesswork when deciding which model or engine fits your Mac. Repeated runs and benchmark history make results easier to compare over time.

Command

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add commands/druide67/asiai/bench.pt
Clone the repo
git clone --depth 1 https://github.com/druide67/asiai

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for bench.pt

README.md
[![agentmods](https://agentmods.dev/badge/commands/druide67/asiai/bench.pt.svg)](https://agentmods.dev/commands/druide67/asiai/bench.pt)
Your own site
<a href="https://agentmods.dev/commands/druide67/asiai/bench.pt"><img src="https://agentmods.dev/badge/commands/druide67/asiai/bench.pt.svg" alt="Measured on agentmods" height="20"></a>
Per session 29 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,767 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00029 $0.01767
Opus 5 $0.00015 $0.00883
Sonnet 5 $0.00006 $0.00353
Haiku 4.5 $0.00003 $0.00177

Measured 4d ago against content hash 84b7304354c0, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

bench.pt scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Asks for rootmediumPrivilege escalation

A mod that escalates privileges can change anything on the machine, not only the project.

| `--power` | Validação cruzada de energia com sudo powermetrics (IOReport sempre ativo) |
docs/commands/bench.pt.md · 170 lines

How it starts

The opening of the file, as written. The whole thing — 170 lines — stays where its author put it; the contents beside it link to each section on GitHub.

asiai bench

Benchmark cross-engine com prompts padronizados.

Uso

asiai bench [options]

Opções

Opção Descrição
-m, --model MODEL Modelo para benchmark (padrão: auto-detecção)
-e, --engines LIST Filtrar motores (ex: ollama,lmstudio,mlxlm)
-p, --prompts LIST Tipos de prompt: code, tool_call, reasoning, long_gen
-r, --runs N Execuções por prompt (padrão: 3, para mediana + desvio padrão)
--power Validação cruzada de energia com sudo powermetrics (IOReport sempre ativo)
--context-size SIZE Prompt de preenchimento de contexto: 4k, 16k, 32k, 64k
--export FILE Exportar resultados para arquivo JSON
-H, --history PERIOD Mostrar benchmarks anteriores (ex: 7d, 24h)
-Q, --quick Benchmark rápido: 1 prompt (code), 1 execução (~15 segundos)
--compare MODEL [MODEL...] Comparação cross-model (2-8 modelos, mutuamente exclusivo com -m)
--card Gerar benchmark card compartilhável (SVG local, PNG com --share)
--share Compartilhar resultados no banco de dados de benchmark da comunidade

Exemplo

asiai bench -m qwen3.5 --runs 3 --power
  Mac Mini M4 Pro — Apple M4 Pro  RAM: 64.0 GB (42% used)  Pressure: normal

Benchmark: qwen3.5

  Engine       tok/s (±stddev)    Tokens   Duration     TTFT       VRAM    Thermal
  ────────── ───────────────── ───────── ────────── ──────── ────────── ──────────
  lmstudio    72.6 ± 0.0 (stable)   435    6.20s    0.28s        —    nominal
  ollama      30.4 ± 0.1 (stable)   448   15.28s    0.25s   26.0 GB   nominal

  Winner: lmstudio (2.4x faster)
  Power: lmstudio 13.2W (5.52 tok/s/W) — ollama 16.0W (1.89 tok/s/W)

Prompts

Quatro prompts padronizados testam diferentes padrões de geração:

Nome Tokens Testa
code 512 Geração de código estruturado (BST em Python)
tool_call 256 Chamada de função JSON / seguimento de instruções
reasoning 384 Problema matemático de múltiplas etapas
long_gen 1024 Throughput sustentado (script bash)

Read the full file on GitHub · 170 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 4d ago First seen · 170 lines · 29 tokens per session scan B 84b7304354c0

Subscribe to this mod's changes

bench.pt is a command published in the GitHub repository druide67/asiai (11 stars, last pushed 5d ago), licensed Apache-2.0. It adds 29 tokens to every session and 1,767 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it B with 1 finding (asks for root). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.