DocOps (Codex, Skills): leaderboard
DocOps (2026) evaluates agents on 210 editing and workflow tasks over native Excel, Word, PowerPoint and PDF files, each checked by a deterministic verifier of structure, semantics and preservation of untouched content. This board runs the models through the Codex agent harness with the Anthropic document skills; the score is the percent of tasks passed, higher is better.
Metric: Task pass rate (%; 210 native-format document tasks in Excel, Word, PowerPoint and PDF (50 atomic edits, 40 compositional, 60 single-document and 60 cross-document workflows), passed only when the submitted artifact satisfies every structural, semantic and preservation predicate of a deterministic in-container verifier; run through Harbor; Codex harness with the Anthropic xlsx, docx, pptx and pdf skills; higher is better). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 67.1 | #49 |
| 2 | GPT-5.4 | 66.2 | #123 |
| 3 | DeepSeek V4 Pro | 42.4 | #188 |
| 4 | Gemma 4 31B | 33.8 | #394 |
| 5 | Qwen 3.5 27B | 29.5 | #359 |
| 6 | Qwen 3.5 122B A10B | 28.1 | #331 |
| 7 | Qwen 3.6 35B A3B | 21.9 | #462 |
| 8 | Qwen 3.6 27B | 21 | #341 |
| 9 | Qwen 3.5 35B A3B | 17.6 | #482 |
| 10 | Qwen 3.5 9B | 8.6 | #803 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=docops-codex-skills · How It Works · Data refreshed daily, snapshot 2026-11-13.