Artificial intelligence firm Anthropic has announced the launch of Claude Opus 4.7, the latest iteration in its series of hybrid reasoning models, marking another rapid product release in 2026.

The company described Opus 4.7 as its most advanced publicly available AI system to date, offering significant improvements in multi-step reasoning and coding capabilities. However, it noted that the model has been deliberately designed to be less powerful than Claude Mythos, which remains restricted due to safety concerns.

According to a company statement, the new model is immediately accessible through the Claude AI platform, its official application programming interface (API), and enterprise partners including Microsoft Foundry. While pricing remains unchanged from the previous version, users may experience increased token usage due to the system’s deeper processing requirements.

Anthropic said the model’s enhanced reasoning leads to more extensive output generation, which could raise operational costs. To address this, the company has released a migration guide outlining strategies to optimise token consumption and manage expenses.

In a blog post, Anthropic highlighted improvements in visual intelligence, document analysis and complex coding tasks, stating that the system demonstrates greater refinement and creativity in handling professional workloads. It added that the model is capable of producing higher-quality interfaces, presentations and documents.

Early users have reported strong performance in executing complex programming assignments, noting that the system maintains consistency over long-running tasks. The model is also said to closely follow user instructions and employ internal checks to validate its outputs.

A detailed model card indicates that while Opus 4.7 trails the restricted Claude Mythos, it remains competitive with leading models developed by OpenAI and Google.

Anthropic further claimed that the model achieved leading results among publicly available systems on Humanity’s Last Exam without external tools, though it performed slightly below GPT-5-4-Pro when such tools were used.

The company emphasised that the model carries a low risk profile, reporting notable reductions in severe hallucinations and critical informational gaps. It added that the system demonstrates improved reliability and accuracy compared to its predecessors in the rapidly evolving artificial intelligence sector.