FORTE - Avg@3: leaderboard
FORTE (Full-cycle Office Real-world Task Evaluation, AGI-Eval): office tasks across 15 corporate professions with real xlsx, docx, pptx, pdf and image files, solved in the OpenClaw agent runtime and graded all-or-nothing against expert rubrics by an LLM judge. Avg@3 is mean task success over three runs, thinking enabled; %, higher is better.
Metric: Avg@3 (%, mean task success over three runs). Source: agi-eval-official.github.io. Saturation forecast: Around February 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 77.78 | #33 |
| 2 | Claude Opus 4.7 | 77.6 | #51 |
| 3 | Claude Opus 4.8 | 77.23 | #39 |
| 4 | GLM-5.2 | 76.87 | #80 |
| 5 | GLM-5.1 | 73.77 | #109 |
| 6 | Claude Opus 4.6 | 73.22 | #67 |
| 7 | LongCat 2.0 | 73.22 | #220 |
| 8 | MiniMax-M3 | 71.22 | #151 |
| 9 | DeepSeek V4 Flash | 70.67 | #155 |
| 10 | DeepSeek V4 Pro | 70.49 | #112 |
| 11 | Gemini 3.1 Pro (Preview) | 70.31 | #64 |
| 12 | GPT-5.4 | 68.85 | #83 |
| 13 | Claude Sonnet 4.6 | 68.31 | #94 |
| 14 | Kimi K2.6 | 63.93 | #113 |
| 15 | Qwen 3.6 Plus | 62.48 | #124 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=forte-avg-3 · How It Works · Data refreshed daily, snapshot 2026-11-05.