VECTOR WIREAI INTELLIGENCE
UTC
Refresh Models Deals Regulatory Sources

InternVL-3-78B vs InternVL-3-8B

26 SHARED BENCHMARKS

Across 26 shared benchmarks, InternVL-3-78B scores higher on 11 and InternVL-3-8B on 14, with 1 level. The widest gap is HallusionBench, where InternVL-3-8B scores 49.2 against 40.2.

OPENGVLABVSOPENGVLAB26 SHARED1114 HEAD-TO-HEAD
A-Bench_VAL75.975.9
AI2D83.585.1
BLINK51.955.9
CCBench70.877.8
ChartQAPro44.437.3
InHouse Dataset A41.540.6
InHouse Dataset B42.636.3
Mathverse49.343.7
mathvision34.829.6
MathVista70.169.5
MMStar66.168.4
MMT-Bench_VAL63.765.2
MTVQA_TEST27.630.3
OCRVQA_TEST35.639
POPE88.990.6
Q-Bench1_VAL7876
RealWorldQA74.371.1
ScienceQA_TEST97.298
ScienceQA_VAL95.197.8
SEEDBench_IMG77.577
SEEDBench2_Plus68.569.5
TextVQA-val83.582.2

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.