Anthropic Releases Claude Opus 4.7, Retaking AI Lead in Tight Race Against OpenAI and Google

Anthropic has publicly released Claude Opus 4.7, its most powerful large language model to date, narrowly retaking the lead in the fiercely competitive race for the most capable generally available AI. The release comes as the company continues to restrict access to an even more powerful successor, codenamed Mythos, which is undergoing cybersecurity testing with a select group of enterprise partners to patch vulnerabilities it rapidly exposed in their software.

Opus 4.7 now surpasses its most direct rivals—OpenAI's GPT-5.4, released in early March 2026, and Google's Gemini 3.1 Pro from February—on key benchmarks for agentic coding, scaled tool-use, agentic computer use, and financial analysis. However, the margins are razor-thin. On directly comparable benchmarks, Opus 4.7 leads GPT-5.4 by a slim 7-4 margin. It currently tops the GDPVal-AA knowledge work evaluation with an Elo score of 1753, ahead of GPT-5.4's 1674 and Gemini 3.1 Pro's 1314.

The launch underscores the escalating intensity of the frontier model race, where incremental gains are becoming harder to secure. More significantly, it highlights a strategic divergence in deployment. While Opus 4.7 is publicly available, the existence of the restricted Mythos model points to a new phase of controlled, security-focused testing for the most advanced capabilities. This two-track approach signals growing pressure on AI labs to balance rapid public iteration with the immense security and stability risks posed by their most powerful systems before wider release.