PI OfficeQA Pro v2
Professional Work · 2026-11-03
Prime Intellect's repaired OfficeQA Pro v2, reported November 3, 2026. It changes the default prompt, expands the corpus and drops some tasks so runs need no web access, and replaces the upstream regex-and-Python grading script with a reference-answer judge. The post calls it on the verge of saturation and doesn't state its task count or which tasks it dropped, so scores aren't comparable with officeqa-pro-v2-reported or officeqa-pro-card.
Benchmark source: Primeintellect
Top models (higher is better)
| Model | Score | Source |
|---|---|---|
| Claude Opus 5.5 | 85.6 | Prime Intellect, Learnings from Auditing Public Evaluations · Prime Intellect |
| GPT-6.1 Sol | 81.1 | Prime Intellect, Learnings from Auditing Public Evaluations · Prime Intellect |
| DeepSeek V4.1 Flash 1017 | 78.8 | Prime Intellect, Learnings from Auditing Public Evaluations · Prime Intellect |