The company said Opus 4.7 improves on Opus 4.6 in difficult, long-running coding tasks, with better instruction-following, greater consistency and stronger self-checking before returning answers. Anthropic also said the model is available from Thursday across its Claude products and API as well as through 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.
That release strategy matters because it comes only about a week after Anthropic drew attention for sharply limiting Mythos Preview, a model it has described as a step up in cyber capability. Under Project Glasswing, the company has said Mythos Preview is being handled as a restricted frontier system after internal testing suggested it could outperform all but the most skilled human experts at finding and exploiting software vulnerabilities. Anthropic’s choice to advance Opus while keeping Mythos behind the fence suggests it is trying to widen adoption without unleashing its most sensitive capabilities into the open market.
Pressure from customers helps explain the timing. Anthropic has increasingly positioned Claude as an enterprise-grade assistant for coding, research and knowledge work, and Opus 4.7 is framed as a practical upgrade for those workloads rather than a dramatic public debut of its most powerful model. On its launch page, the company highlighted testing and endorsements from software and enterprise users including Cursor, Notion, Databricks, Vercel, Ramp and others, all pointing to better results on complex engineering tasks, fewer tool errors and stronger reliability over long sessions.
The sharper focus on coding also reflects a broader shift in the AI market. Model makers are no longer competing only on chatbot fluency; they are fighting to become the default system for software teams, analysts and corporate agents that can carry out multi-step work with limited supervision. Anthropic explicitly says Opus 4.7 can take on harder coding work that once required closer oversight, and it has added a new “xhigh” effort setting to let users trade speed for deeper reasoning on difficult tasks. The company also warned that the model’s improved instruction-following may force developers to retune older prompts and harnesses because it interprets directions more literally than earlier versions.
Safety remains central to the message, though not without caveats. Anthropic said Opus 4.7 shows a broadly similar safety profile to Opus 4.6, with improvements in honesty and resistance to prompt-injection attacks, while acknowledging weaker performance on some measures, including a tendency to provide overly detailed harm-reduction advice on controlled substances. It also said Mythos Preview still appears better aligned by some of its internal evaluations, a notable admission given that Mythos remains unreleased to the public.
Outside the company, concern about Mythos has already spilled into finance and policy circles. Reuters reported on April 9 that U. S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened bank chief executives to warn them about cyber risks tied to Anthropic’s latest model. Reuters also reported that access to Mythos would be limited to about 40 technology companies, underscoring how unusual the restrictions are for a major AI launch and how seriously officials appear to be taking the possibility of AI-assisted cyber exploitation.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
