Jina VLM Release - GSM8K: leaderboard
Jina-VLM technical report2512.04032v1; native jina-vlm text-only performance measured using the authors’ evaluation code, distinct from its Qwen3-1.7B-Base initialization. Exact few-shot and output-budget settings are not specified. GSM8K. Native printed score retained on its declared scale.
Metric: Accuracy (%). Source: arxiv.org. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | jina-vlm | 71.3 |
Interactive version: theaggregate.ai/benchmark?slug=jina-vlm-release-gsm8k · How It Works · Data refreshed daily, snapshot 2026-11-05.