Anthropic Releases Claude Opus 4.7 to Remind Everyone How Great Mythos Is
New model scores 64.3% on SWE-bench Pro but is 'less broadly capable' than unreleased Mythos Preview.
Anthropic has launched Claude Opus 4.7, positioning it as a significant but targeted upgrade over its predecessor. The new model delivers improved performance in key professional areas, most notably retaking the lead in agentic coding among publicly available models with a score of 64.3% on the SWE-bench Pro and Verified benchmarks. It also shows advancements in autonomously navigating operating systems (agentic computer use) and handling graduate-level reasoning. However, in a notable trade-off, its cybersecurity vulnerability reproduction score dipped slightly to 73.1% from 4.6's 73.8%, which Anthropic attributes to new, automated safeguards designed to block high-risk cyber requests.
Despite these improvements, the release narrative is heavily framed by the shadow of Claude Mythos Preview, an unreleased model deemed too powerful for general availability. Anthropic's announcement repeatedly compares Opus 4.7 to Mythos, explicitly calling it "less broadly capable" and positioning it as a testbed for new safety features before they are implemented in the more advanced model. The benchmarking data shared shows Mythos Preview outperforming all other major models, making Opus 4.7's launch feel like a promotional vehicle for Anthropic's cutting-edge, restricted technology. The new model is available immediately across all Claude platforms and APIs at no price change.
- Claude Opus 4.7 scores 64.3% on SWE-bench Pro, retaking the public lead for agentic coding.
- Cybersecurity performance dipped to 73.1% due to new automated safeguards blocking high-risk requests.
- Anthropic explicitly frames it as 'less broadly capable' than its unreleased, benchmark-dominating Claude Mythos Preview.
Why It Matters
Professionals get a safer, more capable coding assistant, but the hype signals a much more powerful, restricted AI tier is coming.