List your company

Visual Reasoning on Tertis

44.67AccuracyCoLVR
20.396426.69823339.3018May 9, 2026

Evaluation results

MethodLinks
CoLVR2026.0544.67
Mirage2026.0543.33
Qwen2.5-VL-7BTraining Paradigm=vanilla SFT2026.0542.67
Qwen2.5-VL-7BTraining Paradigm=vanilla SFT + GRPO2026.0541.33
Gemini-2.5-flash2026.0531.33
GPT-4o2026.0530.67
Qwen2.5-VL-7B2026.0524
InternVL3.5-8B2026.0522
Qwen2.5-VL-7BReasoning Strategy=CoT2026.0521.33