Model6h ago
Claude Opus 5 Dominates ARC-AGI-3 Benchmark
A 30.2 percent score on ARC-AGI-3 puts Anthropic's Opus 5 far ahead of GPT-5.6 Sol's 7.8 percent. The model independently formulated reflection equations to solve complex logic puzzles. This behavior suggests a leap in reasoning capabilities. Practitioners can now expect significantly better performance on tasks requiring abstract logic and novel problem solving.