InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

Benchmark Model Rank Results
video-question-answering-on-mvbenchInternVL3-78B#1Avg.: 78.7
visual-question-answering-on-mm-vetInternVL3-78B#1GPT-4 score: 81.3
visual-question-answering-vqa-on-ai2d-test-with-maskInternVL3-78B#1Accuracy: 89.7
visual-question-answering-vqa-on-mmeInternVL3-78B#1Score: 2549.8
visual-question-answering-vqa-on-mmt-benchInternVL3-78B#1Accuracy: 73.2
visual-question-answering-vqa-on-mtvqaInternVL3-78B#2Accuracy: 32.5
zero-shot-video-question-answer-on-longvideobenchInternVL3-78B#3Accuracy (% ): 65.7
visual-question-answering-vqa-on-infographicvqaInternVL3-78B#4ANLS: 86.5
visual-question-answering-vqa-on-mmbench-cnInternVL3-78B#4Accuracy: 88.7
visual-grounding-on-refcoco-refcoco-refcocog-overallInternVL3-78B#5Accuracy: 91.4
optical-character-recognition-ocr-on-ocrbenchInternVL3-78B#6Score: 906
visual-question-answering-vqa-on-mmbench-enInternVL3-78B#7Accuracy: 89.0
chart-question-answering-on-charxiv-dqInternVL3-78B#8Accuracy: 85.1
common-sense-reasoning-on-winograndeInternVL3-78B#8Accuracy: 87.8
visual-question-answering-on-docvqa-testInternVL3-78B#8ANLS: 0.954
visual-question-answering-vqa-on-textvqa-valInternVL3-78B#9Accuracy: 84.3
vision-language-understanding-on-mmbench-v1-1InternVL3-78B#10Accuracy: 87.7
chart-question-answering-on-chartqaInternVL3-78B#111:1 Accuracy: 89.7
video-question-answering-on-mlvuInternVL3-78B#12Accuracy: 79.5
multi-task-language-understanding-on-c-evalInternVL3-78B#14Accuracy: 89.5
visual-question-answering-vqa-on-blinkInternVL3-78B#15Accuracy: 66.3
video-question-answering-on-video-mme-with-subtitlesInternVL3-78B (w/ subtitles)#16Accuracy: 75.7
visual-question-answering-vqa-on-hallusionbench-average-accuracyInternVL3-78B#16Accuracy: 59.1
visual-question-answering-vqa-on-realworldqaInternVL3-78B#16Accuracy: 78.0
zero-shot-video-question-answer-on-video-mme-w-o-subsInternVL3-78B#16Accuracy (%): 72.7
math-word-problem-solving-on-mathInternVL3-78B#19Accuracy: 78.9
multimodal-reasoning-on-mmmuInternVL3-78B#22Accuracy: 72.2
chart-question-answering-on-charxiv-rqInternVL3-78B#25Accuracy: 46.0
vision-language-understanding-on-mmstarInternVL3-78B#31Accuracy: 72.5
question-answering-on-natural-questionsInternVL3-78B#33EM: 39.0
multimodal-reasoning-on-mathvistaInternVL3-78B#34Accuracy: 79.0
multimodal-reasoning-on-mathvisionInternVL3-78B#38Accuracy: 43.1
code-generation-on-mbppInternVL3-78B#40Accuracy: 76.7
code-generation-on-humanevalInternVL3-78B#52Pass@1: 82.3
arithmetic-reasoning-on-gsm8kInternVL3-78B#68Accuracy: 90.5