In 2026, Anthropic has been swiftly launching products and garnering attention, and on Thursday, the AI firm introduced Claude Opus 4.7. This version is the most sophisticated available to the public, although Anthropic pointed out that it does not match the power of Claude Mythos, which was considered too hazardous for general release. The Claude Opus series is recognized for its hybrid reasoning models featuring multi-step reasoning and enhanced coding abilities. Prior to the reveal of Claude Mythos on April 7, Claude Opus held the title of Anthropic’s most advanced AI series.

Claude Opus 4.7 is now available through Claude AI, the Claude API, and partners such as Microsoft Foundry, maintaining the same price point as its predecessor, Claude Opus 4.6. However, it consumes more output tokens due to its increased effort levels. Users can access optimization recommendations in the Opus 4.7 migration guide.

Claude Opus 4.7 brings enhancements compared to 4.6, especially in advanced coding, visual intelligence, and document analysis. It demonstrates greater creativity and yields higher-quality outcomes for professional endeavors. Users have reported successfully assigning complex coding assignments to Opus 4.7, which manages them with accuracy and consistency.

Anthropic released a model card that compares Claude Opus 4.7 with other models, indicating it falls short of the unreleased Claude Mythos on benchmarks like Humanity’s Last Exam. Opus 4.7 achieved a score of 46.9 percent on HLE, exceeding other frontier models except for Claude Mythos. With the tools available, GPT-5-4-Pro surpassed Opus 4.7, while Mythos outperformed both.

Opus 4.7 exhibits a low probability of misaligned behaviors, akin to Opus 4.6, with a decrease in hallucinations and reward hacking. It is more consistently honest, demonstrating fewer omissions and enhanced factual accuracy. Anthropic’s model card offers comprehensive benchmark results.