Anthropic Releases Claude Opus 4.7: Its Most Capable Public Model Yet

Anthropic has officially released Claude Opus 4.7, the company's newest publicly available flagship AI model. Announced on April 16, 2026, the model is rolling out across Claude.ai, the API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

Anthropic describes Opus 4.7 as its most capable generally available model to date, with meaningful improvements on SWE-bench Verified, SWE-bench Pro, and GPQA Diamond versus its predecessor, Opus 4.6. Third-party trackers like Vellum AI and VentureBeat cite scores in the high-80s on SWE-bench Verified and mid-60s on SWE-bench Pro, placing it firmly at the top of coding benchmarks.

The model brings a significant visual reasoning jump from 69.1% to 82.1%, and now supports images up to 2,576 pixels. A new "xhigh" effort level has been introduced for tasks requiring maximum reasoning depth. Despite the leap in capability, Anthropic openly noted that Opus 4.7 still trails their restricted-access Claude Mythos model in raw capability, positioning Opus 4.7 as the "safe, production-ready" frontier option.

The release triggered immediate industry integrations. Perplexity, for example, has already made Claude Opus 4.7 its default orchestration model for complex queries. On LM Arena, Claude Opus 4.6 Thinking currently holds the top text slot around 1,502 Elo, with votes on Opus 4.7 now accumulating.