InternVL2-8B: benchmark results

OpenGVLab's 8B vision-language model (July 2024) pairing InternViT-300M with internlm2_5-7b-chat for image, document and video understanding. Provider: Shanghai AI Lab. Released 2024-07-04. Access: Open.

Aggregate Rating 1487 ± 6, rank #817 of 1656 rated models, from 1343 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Open LMM Reasoning - MMMath - Difficulty-easy45.4Accuracy (%)100
Open LMM Reasoning - MMMath - Knowledge-L1-Properties of Shapes29.4Accuracy (%)100
Open LMM Reasoning - MMMath - Knowledge-L2-Acute Angle Trigonometric Functions19.7Accuracy (%)100
Open LMM Reasoning - MMMath - Knowledge-L2-Circle34.3Accuracy (%)100
Open LMM Reasoning - MMMath - Knowledge-L2-Intersecting and Parallel Lines33.8Accuracy (%)100
Open LMM Reasoning - MMMath - Knowledge-L2-Symmetry of Shapes11.9Accuracy (%)100
SayNext-Bench (IEMOCAP) - BERTScore50.16BERTScore F1 with deberta-xlarge-mnli (times 100, 0-100) of 100
SayNext-Bench (IEMOCAP) - Sentence-BERT Similarity0.18Mean Sentence-BERT cosine similarity (-1 to 1) between the p100
SayNext-Bench (PC19K) - BERTScore53.87BERTScore F1 with deberta-xlarge-mnli (times 100, 0-100) of 100
SayNext-Bench (PC19K) - Sentence-BERT Similarity0.47Mean Sentence-BERT cosine similarity (-1 to 1) between the p100
SayNext-Bench (PC2K) - Sentence-BERT Similarity0.45Mean Sentence-BERT cosine similarity (-1 to 1) between the p100
OpenVLM MMT-Bench - Disease Diagnosis100Score (%)99.8

Interactive version: theaggregate.ai/model?slug=internvl2-8b · How It Works · Data refreshed daily, snapshot 2026-10-20.