ZoomGainBench - Chinese Counting: leaderboard
ZoomGainBench - Chinese Counting: Fine-grained perception accuracy (%). ZoomGain arXiv 2611.14076v1: ZoomGainBench bilingual fine-grained perception scores, weighted within-language by category counts. Table 3 author experiments train Qwen3.5-4B and 9B on 9K ZoomGain samples; official inclusionAI release identifies them as ZoomGain-4B/9B. ZwZ-retrained controls do not identify released checkpoints and are held. Table 3 Gemini-3-Flash (51.27) conflicts with Table 2 Gemini-3.6-Flash; that row is held, never inferred equivalent. Qwen3-VL labels are held absent Instruct/Thinking binding. Other source-named released models retain their exact labels and unreported effort. Appendix F Table 2 settings are retained per observation: native temperature, output budgets, thinking flags or question suffix, and agentic turn/tool-call limits. Provider-default temperature and thinking remain unspecified where the source gives no value.
Metric: Fine-grained perception accuracy (%). Source: arxiv.org. Saturation forecast: Around August 2028. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 27B (Non-reasoning) | 70.21 | #339 (Qwen 3.6 27B) |
| 2 | Kimi K3 (Thinking) | 64.89 | #53 (Kimi K3) |
| 3 | GPT-5.5 | 63.83 | #48 |
| 4 | GPT-5.6 Sol | 63.83 | #29 |
| 5 | Claude Opus 5 | 62.77 | #20 |
| 6 | Qwen 3.8 27B (Non-reasoning) | 62.77 | #199 (Qwen 3.8 27B) |
| 7 | Qwen 3.5 397B A17B (Non-reasoning) | 62.77 | #242 (Qwen 3.5 397B A17B) |
| 8 | Qwen 3.5 27B (Non-reasoning) | 58.51 | #358 (Qwen 3.5 27B) |
| 9 | Qwen 3.6 35B A3B (Non-reasoning) | 52.13 | #461 (Qwen 3.6 35B A3B) |
| 10 | Qwen 3.5 35B A3B (Non-reasoning) | 51.06 | #480 (Qwen 3.5 35B A3B) |
| 11 | Qwen 3.5 9B (Non-reasoning) | 47.87 | #803 (Qwen 3.5 9B) |
| 12 | Qwen 3.5 4B (Non-reasoning) | 46.81 | #1180 (Qwen 3.5 4B) |
| 13 | Gemini 3.6 Flash | 45.74 | #72 |
| 14 | GLM-4.6V (Non-reasoning) | 36.17 | #698 (GLM-4.6V) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=zoomgainbench-chinese-counting · How It Works · Data refreshed daily, snapshot 2026-11-12.