GPT-5.5 vs Claude Opus 4.7: Claude Retains 6-Point Lead on SWE-bench Pro (64.3% vs 58.6%)

Lushbinary·medium signal

Despite GPT-5.5's strong launch, independent benchmarks show Claude Opus 4.7 maintains a significant lead on SWE-bench Pro — the multi-file GitHub issue resolution benchmark — at 64.3% vs 58.6%. GPT-5.5 leads on Terminal-Bench 2.0 (82.7%) and general autonomy tasks, but for the complex real-world coding that matters most to professional developers, Anthropic's model still outperforms. This creates a split narrative: OpenAI leads on agentic breadth, Anthropic on coding depth.