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.
npx agentmods add commands/druide67/asiai/bench.kogit clone --depth 1 https://github.com/druide67/asiaiWrote 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.
[](https://agentmods.dev/commands/druide67/asiai/bench.ko)<a href="https://agentmods.dev/commands/druide67/asiai/bench.ko"><img src="https://agentmods.dev/badge/commands/druide67/asiai/bench.ko.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00037 | $0.01937 |
| Opus 5 | $0.00018 | $0.00968 |
| Sonnet 5 | $0.00007 | $0.00387 |
| Haiku 4.5 | $0.00004 | $0.00194 |
Grade B, and why
bench.ko 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` | sudo powermetrics로 전력 교차 검증 (IOReport는 항상 활성) | This is a copy
100% identical to bench.ja — 150 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
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
표준화 프롬프트를 사용한 교차 엔진 벤치마크.
사용법
asiai bench [options]
옵션
| 옵션 | 설명 |
|---|---|
-m, --model MODEL |
벤치마크할 모델 (기본: 자동 감지) |
-e, --engines LIST |
엔진 필터 (예: ollama,lmstudio,mlxlm) |
-p, --prompts LIST |
프롬프트 유형: code, tool_call, reasoning, long_gen |
-r, --runs N |
프롬프트당 실행 횟수 (기본: 3, 중앙값 + 표준편차용) |
--power |
sudo powermetrics로 전력 교차 검증 (IOReport는 항상 활성) |
--context-size SIZE |
컨텍스트 필 프롬프트: 4k, 16k, 32k, 64k |
--export FILE |
결과를 JSON 파일로 내보내기 |
-H, --history PERIOD |
과거 벤치마크 표시 (예: 7d, 24h) |
-Q, --quick |
빠른 벤치마크: 1 프롬프트(code), 1회 실행 (~15초) |
--compare MODEL [MODEL...] |
교차 모델 비교 (2-8 모델, -m과 배타적) |
--card |
공유 가능한 벤치마크 카드 생성 (로컬 SVG, --share로 PNG) |
--share |
커뮤니티 벤치마크 데이터베이스에 결과 공유 |
예시
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)
프롬프트
4개의 표준화 프롬프트가 다른 생성 패턴을 테스트합니다:
| 이름 | 토큰 | 테스트 내용 |
|---|---|---|
code |
512 | 구조화된 코드 생성 (Python BST) |
tool_call |
256 | JSON 함수 호출 / 지시 따르기 |
reasoning |
384 | 다단계 수학 문제 |
long_gen |
1024 | 지속 처리량 (bash 스크립트) |
--context-size를 사용하면 대규모 컨텍스트 필 프롬프트로 테스트할 수 있습니다.
교차 엔진 모델 매칭
러너는 엔진 간 모델 이름을 자동 해석합니다 — gemma2:9b(Ollama)와 gemma-2-9b(LM Studio)는 같은 모델로 매칭됩니다.
JSON 내보내기
결과를 공유 및 분석용으로 내보내기:
asiai bench -m qwen3.5 --export bench.json
JSON에는 머신 메타데이터, 엔진별 통계(중앙값, CI 95%, P50/P90/P99), 원시 실행별 데이터, 전방 호환성을 위한 스키마 버전이 포함됩니다.
리그레션 감지
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.
- 4d ago First seen · 170 lines · 37 tokens per session scan B cfe0938bbd59
bench.ko is a command published in the GitHub repository druide67/asiai (11 stars, last pushed 5d ago), licensed Apache-2.0. It adds 37 tokens to every session and 1,937 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it B with 1 finding (asks for root). It is 100% identical to bench.ja, differing in 150 lines, and is treated as a copy.
Other commands, from other repositories
perfup
Autonomous performance optimization: research, PoC, benchmark, implement, review, PR.
integrate-pipeline
Integrate a new document parsing pipeline into ParseBench: $ARGUMENTS.
netllm-connect
Wire Cursor, Claude Code, Codex, or Honcho to the local netllm router.
netllm-setup
First-time netllm install from this repo (uv sync, init, discover, verify).
netllm-swarm
Configure multi-machine LAN mesh for netllm (mDNS, peers, gateway).
/opsx-explore
Enter explore mode - think through ideas, investigate problems, clarify requirements.