Share your thoughts, 1 month free Claude Pro on usSee more

Multimodal Understanding on MMBench EN v1.1

89.5Accuracy

Pixelis (Qwen3-VL-8B-Instruct)

Updated 8d ago

Evaluation Results

Method	Links
Pixelis (Qwen3-VL-8B-Instruct) 2026.03		89.5
Pixelis: SFT + RFT 2026.03		88.6
Pixelis: SFT + TTRL 2026.03		88.3
Seed-1.5-VL 2026.03		88
Pixelis: RFT + TTRL 2026.03		87.9
PRM (process reward; tools; 8B) 2026.03		87.8
Pixelis: SFT only 2026.03		87.6
InternVL2.5-78B 2024.12		87.4
Step Self-Consistency (step-level) 2026.03		87.2
Qwen3-VL-30B-A3B-Instruct 2026.03		87
Pixel Reasoner (Qwen3-VL-8B-Instruct) 2026.03		86.9
Gemini-2.5-Pro (minimal) 2026.03		86.6
RA-TTA (retrieval-augmented) 2026.03		86.4
Pixelis: RFT only 2026.03		86.2
Realistic TTA of VLMs (StatA) 2026.03		86.1
RV Self-Consistency (answer-only) 2026.03		86
Qwen2-VL-72B 2024.12		85.9
Pixelis: TTRL only 2026.03		85.9
InternVL2.5-38B 2024.12		85.5
InternVL2-Llama3-76B 2024.12		85.5
InternVL2-40B 2024.12		85.1
LLaVA-OneVision-72B 2024.12		85
Qwen3-VL-8B-Instruct 2026.03		85
InternVL3.5-A3B (with tools) 2026.03		84.8
InternVL2.5-26B 2024.12		84.2
InternVL2.5-8B 2024.12		83.2
GPT-4o-20240513 2024.12		83.1
Vanilla 2024.12		82.6
VisionZip 2024.12		82
iLLaVA 2024.12		81.7
InternVL2-26B 2024.12		81.5
GPT-5 (minimal) 2026.03		81.4
SparseVLM 2024.12		81.2
Claude-3.5-Sonnet 2024.12		80.9
FasterVLM 2024.12		80.8
Qwen2-VL-7B 2024.12		80.7
VisionZip 2024.12		80.7
PyramidDrop 2024.12		80.5
InternVL-Chat-V1.5 2024.12		80.3
GPT-4V 2024.12		80
iLLaVA 2024.12		79.8
iLLaVA 2024.12		79.7
FasterVLM 2024.12		79.6
InternVL2-8B 2024.12		79.5
InternVL2.5-4B 2024.12		79.3
Cambrian-34B 2024.12		78.3
SparseVLM 2024.12		78.3
MiniCPM-V2.6 2024.12		78
PyramidDrop 2024.12		77.7
FasterVLM 2024.12		77.1
VisionZip 2024.12		76.1
InternVL2-4B 2024.12		75.8
SparseVLM 2024.12		75.8
Gemini-1.5-Pro 2024.12		74.6
Qwen2-VL-2B 2024.12		72.2
InternVL2.5-2B 2024.12		72.2
Phi-3.5-Vision-4B 2024.12		72.1
PyramidDrop 2024.12		71.9
InternVL2-2B 2024.12		70.2
InternVL2.5-1B 2024.12		68.4
InternVL2-1B 2024.12		61.6
Claude-3-Opus 2024.12		60.1
LLaVA-OneVision-0.5B 2024.12		59.6