Anthropic made its Claude Opus 4.7 model generally available, citing marked gains in advanced software engineering, long-running autonomy, and higher-resolution vision. Early testers reported improved planning, error recovery, and instruction adherence, with double-digit lifts on coding and agentic benchmarks. The release arrives with stricter cybersecurity guardrails under the company’s Project Glasswing initiative; Anthropic said Opus 4.7’s cyber capabilities are intentionally constrained relative to its higher-end Mythos Preview model, and it invited qualified professionals to a new Cyber Verification Program. Opus 4.7 is available across Anthropic’s products and via Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry, with pricing unchanged at $5 per million input tokens and $25 per million output tokens. New features include a finer-grained “xhigh” effort level, support for higher-resolution images, public-beta task budgets for developers, and new Claude Code tools such as an “/ultrareview” mode and expanded auto permissions. Anthropic said safety performance is broadly similar to Opus 4.6 with some improvements in honesty and resistance to prompt injection, and cautioned that an updated tokenizer and deeper reasoning at higher effort may increase token counts during migration.
Related article:
– Amazon Bedrock: Foundation Models from Leading AI Providers
