ResearchBenchmarksLong-Context Reasoning on GPQA Diamond (out-of-distribution)Follow48.5AccuracyVTC-R125.41231.40637.443.394Jan 29, 2026Evaluation ResultsMethodMethodLinksAccuracyToken ThroughputLatencyVTC-R1Architecture=Qwen3-VL-8BArchitecture=Qwen3-VL-8B2026.0148.59.779.57VTC-R1Architecture=GlyphArchitecture=Glyph2026.014610.736.96SFTArchitecture=GlyphArchitecture=Glyph2026.0138.413.918.35SFTArchitecture=Qwen3-VL-8BArchitecture=Qwen3-VL-8B2026.0137.414.7826.88TokenSkipArchitecture=GlyphArchitecture=Glyph2026.0135.915.459.93Base SFTArchitecture=GlyphArchitecture=Glyph2026.0126.319.7414.43