Proposal and executable sample pack. We build the environments and training tasks around the model, not simply the annotations inside them.
The document below is the Cognyzer proposal for Z.ai: OSWorld compatible computer use environments, task anatomy, quality bar, evidence, and a path from pilot to scale. No pricing in this pack.
Ten tasks. Six Ubuntu RL tasks with scored model rollouts. Two Windows and two macOS tasks as passing human SFT. A run passes only if the artefact left inside the virtual machine satisfies the evaluator.
Full pack, 338 MB. Includes README.md, task.json, human trajectories, Ubuntu model runs, and Windows/macOS seed files.
Download zipUbuntu / Windows / macOS stay in separate folders. Each task has task.json and sft/ with result.txt = 1.000000. Ubuntu also has runs/.
Pack READMEScreenshot only, pyautogui, last 3 screenshots, 3 runs. Pass rate is successes out of three.
| Task | App | Model | Runs | Pass |
|---|---|---|---|---|
| thunderbird-c37b60 | Thunderbird | claude-sonnet-4-5 | 1 1 0 | 0.667 |
| glm-5v-turbo | 0 0 0 | 0.0 | ||
| vlc-5e94af | VLC | claude-sonnet-4-5 | 1 1 0 | 0.667 |
| glm-5v-turbo | 0 0 0 | 0.0 | ||
| vs-code-4b6f82 | VS Code | claude-sonnet-4-5 | 1 0 0 | 0.333 |
| glm-5v-turbo | 0 0 0 | 0.0 | ||
| gimp-ef7602 | GIMP | glm-5v-turbo | 0 0 0 | 0.0 |
| libreoffice-writer-7cfdbc | Writer | glm-5.3-flash | 0 0 0 | 0.0 |
| libreoffice-calc-e4af30 | Calc | glm-5.3-flash | 0 0 0 | 0.0 |
Task definition, seed files, and a passing human trajectory. Not GLM-scored in this pack.
| Folder | OS | Application | Human trajectory |
|---|---|---|---|
| windows/gsheets | Windows | Google Sheets | Reykjavik weather workbook |
| windows/gdocs | Windows | Google Docs | Thesis workshop document edits |
| macos/onlyoffice | macOS | ONLYOFFICE | Fleet insurance audit |
| macos/chrome | macOS | Chrome / Sheets | Q2 ridership derived columns |
Pragnasya S, Co-founder and CTO, Cognyzer, Hyderabad / team@cognyzer.com
Confidential. Shared with Z.ai for evaluation of this collaboration.