Benchmark
xbench-BabyVision
Vision benchmarks measure how well models understand images. Scores are percentages.
All vision benchmarksScores from xbench.
- Models measured
- 12
- Scores recorded
- 12
- Sources
- 1
- IQ weight
- 0.71
Score distribution
- 46.3%
- Mean
- 46.4%
- Median
- 22.2%
- Std dev
- 20.1%
- 10th percentile
- 73.5%
- 90th percentile
- 19.2%
- Min
This benchmark has a weight of 0.71 for chat models, measured over 12 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT-6 Astrapro· xbench | 88.9% |
| 2 | Gemini 3.7 Flash· xbench | 73.5% |
| 3 | Doubao Seed 2.1 Pro· xbench | 66.9% |
| 4 | Qwen3.8 Max Preview· xbench | 65% |
| 5 | Kimi K3· xbench | 49.5% |
| 6 | Claude Fable 5.1· xbench | 46.4% |
| 7 | DeepSeek Flash· xbench | 43.3% |
| 8 | GLM 5.3 Flash· xbench | 36.6% |
| 9 | Grok 4.5· xbench | 25.5% |
| 10 | Inkling FP4· xbench | 21.4% |
| 11 | MiniMax M3· xbench | 20.1% |
| 12 | internvl3-5-241b-a28b· xbench | 19.2% |