ResearchBenchmarksMathematical Reasoning on GSM8K (test) (Accuracy, Token Count, Latency)Mathematical Reasoning on GSM8K (test) (Accuracy, Token Count, Latency)Follow94.7AccuracyVTC-R185.75688.07890.492.722Jan 29, 2026Evaluation resultsMethodMethodLinksAccuracyToken CountLatencyVTC-R1Backbone=Qwen3-VL-8BBackbone=Qwen3-VL-8B2026.0194.71.090.46VTC-R1Backbone=GlyphBackbone=Glyph2026.0193.61.090.34SFTBackbone=Qwen3-VL-8BBackbone=Qwen3-VL-8B2026.0188.11.793.04SFTBackbone=GlyphBackbone=Glyph2026.0187.11.870.93TokenSkipBackbone=GlyphBackbone=Glyph2026.0186.42.251.32Base SFTBackbone=GlyphBackbone=Glyph2026.0186.12.351.38