VISTA-Bench (Pure Text): leaderboard
Metric: Accuracy (%) over the same 1,500 questions given as ordinary text tokens (the control condition of VISTA-Bench; problem images unchanged); zero-shot with VLMEvalKit default decoding, final answers recovered by a GPT-based extractor; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 81.9 | #54 |
| 2 | Qwen 3.5 122B A10B | 79.6 | #170 |
| 3 | GLM-4.6V | 73.9 | #309 |
| 4 | GPT-5.2 | 69.5 | #105 |
| 5 | GLM-4.1V-9B (Thinking) | 67.2 | #457 (GLM-4.1V-9B) |
| 6 | Qwen 3 VL 30B A3B Instruct | 66.7 | #365 |
| 7 | Qwen 3 VL 8B Instruct | 59.1 | #401 |
| 8 | Qwen 2.5 VL 7B Instruct | 54.3 | #643 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=vista-bench-pure-text · How It Works · Data refreshed daily, snapshot 2026-10-11.