--- license: mit language: - en size_categories: - n<1K task_categories: - other tags: - benchmark - computer-use-agent - gui - cli - hybrid-interface - long-horizon - agent-as-judge pretty_name: WeaveBench --- # WeaveBench > A long-horizon, real-world benchmark for computer-use agents with **hybrid GUI + CLI + code** interfaces. πŸ“„ **Paper**: [github.com/weavebench/WeaveBench (paper.pdf)](https://github.com/weavebench/WeaveBench/blob/main/paper.pdf) πŸ’» **Code**: [github.com/weavebench/WeaveBench](https://github.com/weavebench/WeaveBench) 🌐 **Website**: [weavebench.github.io](https://weavebench.github.io) **114 long-horizon, real-world tasks** across 8 work domains, where every task requires the agent to **interleave GUI clicks with shell/code** in one trajectory. Scored by a **trajectory-aware Agent-as-Judge** that reads chat trace + deliverables and emits per-clause evidence β€” much harder to spoof than file-existence checks. Best observed pairing in the paper: **Claude Opus 4.7 + Claude Code at 41.2 % PassRate** β€” far from saturation. This HF repository hosts everything needed to run the benchmark end-to-end: 1. **`tasks/`** β€” 114 paper-final tasks + workspace assets (~207 MB) 2. **`runtime_assets/`** β€” pre-built bootstrap tarballs for the 4 in-VM agent harnesses (~852 MB) 3. **`vm/Ubuntu.qcow2`** β€” the paper-canonical Ubuntu VM image (28.46 GB) 4. **`judge/judge_template.tar.gz`** β€” host-side OpenClaw judge profile template (~7 KB) > πŸ‡¨πŸ‡³ **Users in China**: `export HF_ENDPOINT=https://hf-mirror.com` β€” all `weavebench-download-*` commands honor it automatically. --- ## 1. Layout ``` WeaveBench/ β”œβ”€β”€ tasks/ ← 207 MB β”‚ β”œβ”€β”€ DAV/ DES/ DOC/ DSK/ GAM/ OPS/ SPA/ WEB/ ← 8 domains β”‚ β”‚ └── _task__.md ← 114 task .md files β”‚ └── workspace/// ← 951 supporting files (gt/, exec/, …) β”‚ β”œβ”€β”€ runtime_assets/ ← 852 MB total β”‚ β”œβ”€β”€ openclaw.tar.gz 514 MB (reference harness) β”‚ β”œβ”€β”€ codex.tar.gz 131 MB (OpenAI Codex CLI) β”‚ β”œβ”€β”€ claudecode.tar.gz 72 MB (Anthropic Claude Code) β”‚ β”œβ”€β”€ hermes.tar.gz 127 MB (Nous Research Hermes) β”‚ └── hermes_mcp_wheels.tar.gz 9 MB (offline mcp wheels for Hermes) β”‚ β”œβ”€β”€ vm/ β”‚ └── Ubuntu.qcow2 28.46 GB (v3_eyeson_apps, paper-canonical) β”‚ └── judge/ └── judge_template.tar.gz ~7 KB (OpenClaw judge profile) ``` **Total**: ~29.5 GB. See the [README Β§Resource budget](https://github.com/weavebench/WeaveBench#resource-budget-read-before-you-commit) on the code repo for what to allocate. ## 2. One-command download + run ```bash git clone https://github.com/weavebench/WeaveBench.git && cd WeaveBench export OPENROUTER_API_KEY=sk-or-v1-... # Optional: export HF_ENDPOINT=https://hf-mirror.com # if in China # bash scripts/setup.sh does all four downloads + npm install -g openclaw + judge bootstrap bash scripts/setup.sh # Smoke test (default model: openai/gpt-5.5) weavebench-run \ --harness openclaw \ --model openai/gpt-5.5 \ --tasks_root ./cache/tasks \ --domains WEB --task_filter task_1_ \ --result_dir ./results/smoke ``` Or fetch individual assets: ```bash weavebench-download-dataset --dest ./cache # tasks/ 207 MB weavebench-download-assets --dest ./cache # runtime_assets/ 852 MB total weavebench-download-vm --dest ./cache # vm/Ubuntu.qcow2 28.46 GB weavebench-download-judge # judge/ β†’ ~/judge_agent_test/ ``` ## 3. Try it in 30 seconds (no VM, no API key) If you just want to see what `score.json` from the Agent-as-Judge looks like: ```bash git clone https://github.com/weavebench/WeaveBench.git && cd WeaveBench pip install -e . weavebench-demo ``` This prints a real paper-canonical Claude Opus 4.7 rollout's score with per-artifact / per-clause judge evidence in <1 second β€” no network calls, no tokens. ## 4. Task `.md` schema Each task file has these sections in order: ```markdown # ## Goal β€” short user request (what the agent reads as `instruction`) ## Setup β€” preconditions assumed true in the VM ## Warmup β€” bash commands the orchestrator runs before the agent starts ## Expected Output β€” files the agent must produce in /tmp_workspace/results/ ## Grader β€” Python `def grade(workspace_path, transcript) -> dict` ``` The embedded `grade(...)` function returns `{"score": float ∈ [0, 1], "scores": {sub_rubric: float, ...}, "msg": "..."}` β€” but in the **trajectory-aware judge pipeline** (the only scoring path in `weavebench-run`), this is documentation only. Scoring is done by a host-side OpenClaw judge ([`weavebench/eval/agent_judge`](https://github.com/weavebench/WeaveBench/tree/main/weavebench/eval/agent_judge)) that reads chat trace + deliverables and emits per-clause evidence-based scores. See [docs/AGENT_JUDGE.md](https://github.com/weavebench/WeaveBench/blob/main/docs/AGENT_JUDGE.md). ## 5. Pinned revision for paper reproduction OpenRouter aliases drift over time. To reproduce paper numbers, pin both the dataset SHA and the model snapshot ids β€” see [`docs/REPRODUCE.md`](https://github.com/weavebench/WeaveBench/blob/main/docs/REPRODUCE.md) in the code repo for the canonical SHA, per-table model ids, and per-table commands. ```bash export WEAVEBENCH_DATASET_REVISION= ``` ## 6. Citation ```bibtex @article{li2026weavebench, title = {WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces}, author = {Li, Wanli and Zhou, Bowen and Yang, Yifan and Yu, Yunyao and Li, Dongsheng and Xu, Zhou and Shan, Caihua}, year = {2026}, } ``` ## 7. License - **Tasks**: MIT. - **Runtime tarballs**: each tarball repackages third-party software (Codex CLI is Apache-2.0; Claude Code, Hermes, OpenClaw retain their upstream terms). See [NOTICE](https://github.com/weavebench/WeaveBench/blob/main/NOTICE) in the code repo for full attribution. - **VM image**: derived from OSWorld upstream Ubuntu image (MIT), patched to bake in app snapshots for paper reproducibility β€” full provenance in [docs/REPRODUCE.md](https://github.com/weavebench/WeaveBench/blob/main/docs/REPRODUCE.md).