Pointerbench — Sheets
Multimodal · 2026-07-02
Pointerbench-Sheets is a 500-example synthetic GUI-grounding benchmark for spreadsheet cells, headers, colors, resize handles, grid edges and corners, and relative cell positions. A predicted absolute-pixel click is correct when it lands inside the target bounding box.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 96.0 |
| GPT-5.6 Sol | 92.0 |
| GPT-5.6 Sol Pro | 92.0 |
| Sonnet 4.6 | 82.6 |
| Opus 4.8 | 82.2 |
| Opus 4.7 | 73.8 |
| GPT-5.5 | 72.4 |
| GPT-5.4 | 63.2 |
| GPT-5.6 Luna | 56.0 |
| GPT-5.6 Luna Pro | 54.0 |
| GPT-5.6 Terra Pro | 52.0 |
| GPT-5.6 Terra | 50.0 |
| GPT-5 | 35.2 |
| Kimi K2.5 | 10.8 |
| GPT-5 Mini | 6.0 |
| Qwen3.7-Plus | 5.8 |
| Kimi K2.6 | 5.6 |
| Qwen3.6-Flash-2026-04-16 | 4.2 |
| MiniMax M3 | 3.6 |
| Gemini 3.5 Flash | 3.4 |
| Qwen3.6 Plus (2026-04-02) | 3.0 |
| Qwen3 VL 235B A22B Thinking | 2.6 |
| GPT-5 Nano | 2.0 |
| Qwen3.5-9B | 2.0 |
| Grok 4.3 | 1.2 |
| Kimi K2.7 Code | 0.4 |
| Gemini 3.1 Pro Preview | 0.0 |
| Grok Build 0.1 | 0.0 |
| MiMo-V2.5 | 0.0 |
| Step 3.7 Flash | 0.0 |