Anthropic's Claude Opus 4.7 narrowly edges out OpenAI's GPT-5.4 in benchmark competition, though rivals are closing the gap rapidly.

Claude Opus 4.7 achieves the highest Elo score of 1753 on GDPVal-AA knowledge work evaluation, surpassing GPT-5.4 (1674) and Gemini 3.1 Pro (1314).

The lead is minimal—Opus 4.7 only outperforms GPT-5.4 by 7-4 on directly comparable benchmarks, indicating an increasingly competitive LLM market.

Anthropic is keeping its more powerful successor model, Mythos, restricted to enterprise partners for cybersecurity testing and vulnerability patching.

Opus 4.7 excels in agentic coding, scaled tool-use, agentic computer use, and financial analysis compared to GPT-5.4 (released March 2026) and Gemini 3.1 Pro (February 2026).