{"id":50724343,"url":"https://github.com/drewmattie-code/pipelinescore","last_synced_at":"2026-06-10T03:01:08.817Z","repository":{"id":343545423,"uuid":"1171216151","full_name":"drewmattie-code/pipelinescore","owner":"drewmattie-code","description":"Measure LLM performance on your own equipment. Run a 25-task benchmark against any model with your own API key, get a deterministic score, tier badge, and a place on the public leaderboard.","archived":false,"fork":false,"pushed_at":"2026-06-09T00:11:23.000Z","size":83377,"stargazers_count":0,"open_issues_count":3,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-09T02:12:40.513Z","etag":null,"topics":["ai","ai-evaluation","apache-2","benchmark","claude","hardware-comparison","leaderboard","llama-cpp","llm","lm-studio","local-first","local-llm","local-models","mcp","mlx","ollama","openai","typescript","vllm"],"latest_commit_sha":null,"homepage":"https://pipelinescore.ai","language":"TypeScript","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/drewmattie-code.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":"CODE_OF_CONDUCT.md","threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-03-03T01:47:57.000Z","updated_at":"2026-06-09T00:11:27.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/drewmattie-code/pipelinescore","commit_stats":null,"previous_names":["drewmattie-code/pipelinescore"],"tags_count":2,"template":false,"template_full_name":null,"purl":"pkg:github/drewmattie-code/pipelinescore","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/drewmattie-code%2Fpipelinescore","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/drewmattie-code%2Fpipelinescore/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/drewmattie-code%2Fpipelinescore/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/drewmattie-code%2Fpipelinescore/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/drewmattie-code","download_url":"https://codeload.github.com/drewmattie-code/pipelinescore/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/drewmattie-code%2Fpipelinescore/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34134633,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-10T02:00:07.152Z","response_time":89,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai","ai-evaluation","apache-2","benchmark","claude","hardware-comparison","leaderboard","llama-cpp","llm","lm-studio","local-first","local-llm","local-models","mcp","mlx","ollama","openai","typescript","vllm"],"created_at":"2026-06-10T03:01:07.395Z","updated_at":"2026-06-10T03:01:08.804Z","avatar_url":"https://github.com/drewmattie-code.png","language":"TypeScript","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n\n# PipelineScore\n\n**Benchmark LLMs on YOUR hardware.** Same 34 deterministic tasks, scored entirely on your machine — no judge model, no API key — one 0–100 score. The only public LLM leaderboard that ranks where the model runs — not just which model it is.\n\n[![Live at pipelinescore.ai](https://img.shields.io/badge/live-pipelinescore.ai-0F766E?style=flat-square)](https://pipelinescore.ai)\n[![License: Apache 2.0](https://img.shields.io/badge/license-Apache_2.0-blue?style=flat-square)](LICENSE)\n[![Made with TypeScript](https://img.shields.io/badge/made_with-TypeScript-3178C6?style=flat-square\u0026logo=typescript\u0026logoColor=white)](https://www.typescriptlang.org/)\n[![GitHub stars](https://img.shields.io/github/stars/drewmattie-code/pipelinescore?style=flat-square)](https://github.com/drewmattie-code/pipelinescore/stargazers)\n[![GitHub issues](https://img.shields.io/github/issues/drewmattie-code/pipelinescore?style=flat-square)](https://github.com/drewmattie-code/pipelinescore/issues)\n[![Local-first](https://img.shields.io/badge/local--first-Ollama_·_LM_Studio_·_MLX_·_llama.cpp-0F766E?style=flat-square)](https://pipelinescore.ai/run)\n\n[**🚀 Live at pipelinescore.ai**](https://pipelinescore.ai)\n\n[Live leaderboard](https://pipelinescore.ai/leaderboard/users) · [Methodology](https://pipelinescore.ai/methodology) · [Privacy / BYOK posture](https://pipelinescore.ai/privacy) · [Run the CLI](https://pipelinescore.ai/run)\n\n[![PipelineScore hardware board — every rig ranked by its best score: B200, DGX H100, A100, dual RTX 4090, M-series Macs and more](assets/leaderboard-screenshot.jpg)](https://pipelinescore.ai/leaderboard/hardware)\n\n\u003c/div\u003e\n\n---\n\n## What it looks like\n\n```text\n$ npx @pipelinescore/cli run \\\n    --provider local --endpoint http://localhost:11434 \\\n    --model llama-3.3-70b --hardware-tag m3-max-128gb \\\n    --user your-handle\n\n╭ PipelineScore v0.3.0 ──────────────────╮\n│ Provider:     local                    │\n│ Model:        llama-3.3-70b            │\n│ Hardware:     m3-max-128gb             │\n│ Config tag:   — (base model)           │\n│ User:         your-handle              │\n│ Submit:       yes                      │\n╰────────────────────────────────────────╯\n\nFetched testpack 2026-06-10-v3 from backend.\nRunning 34 tasks ... ████████████████████ 34/34\n\n╭──────────────────── PipelineScore ─────────────────────╮\n│                                                        │\n│   78.4   MAINLINE                                      │\n│   ────                                                 │\n│                                                        │\n│   code ████████░░  79.1     tool_use ██████░░░░  61.4  │\n│   reason ███████░░ 75.8     rag      ████████░░  82.6  │\n│   speed █████░░░░░ 52.3                                │\n│                                                        │\n│   Total tokens: 4,827 · Avg latency: 712ms             │\n│   See your run: pipelinescore.ai/users/your-handle     │\n╰────────────────────────────────────────────────────────╯\n\nOpening your leaderboard page in your browser.\n```\n\n## Quickstart — local model (30 seconds)\n\nIf you have Ollama / LM Studio / MLX / llama.cpp running:\n\n```bash\nnpx @pipelinescore/cli run \\\n  --provider local \\\n  --endpoint http://localhost:11434 \\\n  --model llama-3.3-70b \\\n  --hardware-tag m3-max-128gb \\\n  --user your-handle\n```\n\nSwap port for LM Studio (`1234`), llama.cpp (`8080`), MLX-Omni (`10240`), or LiteLLM proxy (`8000`). Replace `m3-max-128gb` with your rig (`rtx-4090-24gb`, `ryzen-7950x-cpu-only`, `a100-80gb`, anything alphanum + `. _ -`).\n\nThe CLI runs locally, calls your model server, scores the output, and publishes the result to https://pipelinescore.ai/users/your-handle.\n\n## Quickstart — frontier API (BYOK)\n\n```bash\nANTHROPIC_API_KEY=sk-... npx @pipelinescore/cli run \\\n  --provider anthropic --model claude-opus-4-7 \\\n  --user your-handle\n```\n\nOr `--provider openai`. **Your key never reaches our backend** — it goes directly to the provider. See [Privacy](https://pipelinescore.ai/privacy) for the full data-flow.\n\n## Why this leaderboard exists\n\nEvery other ranked LLM list ignores the rig:\n\n| | Hardware-aware? | You can run it yourself? | Local-model coverage | Reproducible | Open source |\n|---|:---:|:---:|:---:|:---:|:---:|\n| **PipelineScore** | ✅ | ✅ | ✅ | ✅ | ✅ Apache 2.0 |\n| LMArena | ❌ | ❌ (preference votes only) | partial | ❌ | partial |\n| Artificial Analysis | ❌ | ❌ (centrally run) | partial | ❌ | ❌ |\n| lm-evaluation-harness | ❌ | ✅ | ✅ | ✅ | ✅ MIT |\n| MMLU / SWE-Bench / TerminalBench | ❌ | ✅ | ✅ | ⚠️ test set leaks fast | ✅ |\n| OpenLLM Leaderboard (HF) | ❌ | ❌ | ✅ | ✅ | ✅ |\n\n**The missing axis is the hardware tag.** Same Llama 4 on an M3 Max vs an RTX 4090 vs an A100 produces three very different real-world experiences. Same RTX 4090 with three different models produces three apples-to-apples comparisons. The benchmark is reproducible, the hardware tag is preserved, the score lands on a public, searchable leaderboard at https://pipelinescore.ai/leaderboard/users.\n\n## Architecture\n\n```mermaid\nflowchart LR\n    A[Your CLI\u003cbr/\u003enpx @pipelinescore/cli] --\u003e|HTTPS\u003cbr/\u003eOpenAI-compat| B[Your model server\u003cbr/\u003eOllama / LM Studio /\u003cbr/\u003eMLX / llama.cpp / vLLM]\n    A --\u003e|HTTPS POST\u003cbr/\u003escore + transcripts| C[api.pipelinescore.ai\u003cbr/\u003eExpress + SQLite\u003cbr/\u003eon Render]\n    C --\u003e|read| D[Cloudflare Worker\u003cbr/\u003eNext.js via OpenNext]\n    D --\u003e|HTTPS GET| E[pipelinescore.ai\u003cbr/\u003epublic leaderboard]\n\n    F[Claude Code skill] --\u003e|invokes| A\n    G[pipelinescore-mcp\u003cbr/\u003eMCP server] --\u003e|invokes| A\n    G --\u003e|reads| C\n\n    style A fill:#0F766E,color:#fff\n    style E fill:#0F766E,color:#fff\n```\n\n**Three integration paths** to drive the CLI:\n1. **Manual** — copy/paste the `npx` command into your terminal\n2. **Skill** — drop [`SKILL.md`](dist/skills/pipelinescore/SKILL.md) into `~/.claude/skills/` and your AI runs it for you\n3. **MCP** — install [`@pipelinescore/mcp`](mcp/) and any MCP-compatible client (Claude Code, Cursor, Codex, Continue, Cline) gets the benchmark as a tool\n\n**Backend never sees your API key.** When `--provider anthropic/openai`, the CLI calls the provider directly. Only the score + transcripts (with API keys stripped) reach our backend. See [SECURITY.md](SECURITY.md) for the full posture.\n\n## The score\n\nFive deterministic categories — code (executed), reason (exact-match), tool use + RAG (JSON-match), speed (measured throughput) — weighted to mirror real LLM usage. One headline number (0–100), category breakdown underneath. Score maps to one of five tiers — TRUNK / MAINLINE / FEEDER / TAP / DRIP — for at-a-glance readability.\n\nFull methodology + weights + anti-cheat: [pipelinescore.ai/methodology](https://pipelinescore.ai/methodology)\n\n## Deeper documentation\n\nThis README is the front door. For specifics:\n\n| | Where |\n|---|---|\n| 🤖 LLM-first usage guide | [AGENTS.md](AGENTS.md) |\n| 🛠️ Local dev setup (backend + web + CLI) | [DEVELOPMENT.md](DEVELOPMENT.md) |\n| 🛡️ BYOK posture + retention policy | [SECURITY.md](SECURITY.md) + [pipelinescore.ai/privacy](https://pipelinescore.ai/privacy) |\n| 🧮 How scores are computed + anti-cheat | [pipelinescore.ai/methodology](https://pipelinescore.ai/methodology) |\n| 🤝 Contributing | [CONTRIBUTING.md](CONTRIBUTING.md) |\n| 📜 Changelog | [CHANGELOG.md](CHANGELOG.md) |\n| 🗣️ Code of conduct | [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md) |\n\n## Contributing\n\nWe need help with:\n- **More benchmark tasks** — submit a PR with a task in `benchmarks/tasks-v1.json`\n- **More local server endpoints** — vLLM, TGI, Ramalama, anything OpenAI-compatible\n- **Hardware tag suggestions** — common rigs we're missing in [seed-local-models.ts](backend/src/seed-local-models.ts)\n- **Bug reports** — file an issue with the failing nickname / model / hardware combo\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for the workflow + [SECURITY.md](SECURITY.md) for the BYOK posture.\n\n## Star History\n\n[![Star History Chart](https://api.star-history.com/svg?repos=drewmattie-code/pipelinescore\u0026type=Date)](https://star-history.com/#drewmattie-code/pipelinescore\u0026Date)\n\nIf this repo is useful to you, a star is the easiest signal to send. It helps surface PipelineScore to other devs running local models.\n\n## License\n\n[Apache 2.0](LICENSE).\n\n## Authors\n\nDrew Mattie · SaaSquach AI Labs (a division of Charles \u0026 Roe Inc.)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdrewmattie-code%2Fpipelinescore","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fdrewmattie-code%2Fpipelinescore","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdrewmattie-code%2Fpipelinescore/lists"}