Claude Opus 4.6 Benchmark Table Showing Dominance Across Agentic Tasks

Description

A detailed benchmark comparison table with a white/light background comparing five AI models: Opus 4.6 (highlighted with a dashed border), Opus 4.5, Sonnet 4.5, Gemini 3 Pro, and GPT-5.2 (all models). The table covers 15 benchmarks including Agentic terminal coding (Terminal-Bench 2.0), Agentic coding (SWE-bench Verified), Agentic computer use (OSWorld), Agentic tool use (tau2-bench), Scaled tool use (MCP Atlas), Agentic search (BrowseComp), Multidisciplinary reasoning (Humanity's Last Exam), Agentic financial analysis (Finance Agent), Office tasks (GDPVal-AA Elo), Novel problem-solving (ARC AGI 2), Graduate-level reasoning (GPQA Diamond), Visual reasoning (MMMU Pro), and Multilingual Q&A (MMMLU). Opus 4.6 leads in most categories with bold scores, particularly strong in agentic tasks like computer use (72.7%), search (84.0%), and novel problem-solving (68.8%). Some cells are highlighted in red/salmon (Opus column leads) or green/cream (competitor leads). Notable results include Gemini 3 Pro leading Visual reasoning without tools at 81.0% and GPT-5.2 leading Graduate-level reasoning at 93.2%