Anthropic released its latest AI model, Claude Opus 4.7, characterized as a “notable improvement” over Opus 4.6, yet “less broadly capable” than the unreleased Opus Mythos Preview. The new model enhances existing strengths, focusing on coding, engineering, and multi-step tasks.
Claude Opus 4.7 shows superior performance in professional knowledge work, claiming to be “more thorough and consistent” in challenging contexts. The model’s benchmarking tests demonstrate its capabilities, with a score of 64.3% in agentic coding on SWE-bench Pro and SWE-bench Verified, reclaiming the top position among publicly available models.
In comparison to Opus 4.6, Opus 4.7 also exhibits improvements in agentic computer use and graduate-level reasoning. However, it shows a slight decrease in cybersecurity vulnerability scores, achieving 73.1% compared to 73.8% for the previous version. Anthropic noted that this change may result from new safeguards intended to detect and block high-risk cybersecurity requests.
The launch of Claude Opus 4.7 appears to promote the Claude Mythos Preview, which has demonstrated superior performance across major benchmarks but is currently available only to select organizations. Anthropic emphasized that Opus 4.7’s cyber capabilities do not match those of Mythos Preview.
“We stated that we would keep Claude Mythos Preview’s release limited and test new cyber safeguards on less capable models first,” the company stated. “Opus 4.7 is the first such model: its cyber capabilities are not as advanced as those of Mythos Preview.”
Claude Opus 4.7 is available immediately across all Claude products and through the company’s API, maintaining the same pricing as previous models.
