Anthropic releases Claude Opus 4.7, a powerful coding model with deliberately reduced cyber capabilities

Anthropic has released Claude Opus 4.7, the company's strongest AI available to the broad market, though not its most powerful overall. At the same time, the company effectively acknowledged that its more advanced model, Mythos Preview, is still too risky for a mass launch because of its strong cyber capabilities.

The new release therefore looks like a compromise between practical power and safety: Opus 4.7 is tuned for complex development, long-running agentic tasks and enterprise work, but ships with additional restrictions on dangerous cyber scenarios. The release came on April 16, 2026, just nine days after the April 7, 2026 announcement of Mythos Preview.

Anthropic describes Mythos as its most powerful AI and, at the same time, the most sensitive one from a safety standpoint. Instead of broad access, the company kept Mythos in the restricted Project Glasswing environment, where selected partners working with critical software and infrastructure are testing the model. The logic is simple: give defenders a head start first, and only then decide whether such a system can be scaled safely to the mass market.

As a product, Opus 4.7 is a direct update to Opus 4.6 with a focus on software engineering.

Anthropic says the model handles the hardest programming tasks better, needs less supervision, follows instructions more precisely and can hold complex multi-step context for longer. Anthropic also stresses that the model more often checks its own output before returning an answer. Visual tasks get an upgrade too: Opus 4.

7 accepts images with a long side of up to 2576 pixels, or roughly 3.75 megapixels, more than three times the limit of earlier Claude models. That matters for agentic scenarios in which the model has to read dense screenshots, diagrams and interfaces.

Anthropic also added a new xhigh mode between high and max, so developers can more precisely choose the balance between reasoning depth and response latency. The model is already available in Claude, through the API, and on Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. Pricing is unchanged from Opus 4.

6: $5 per million input tokens and $25 per million output tokens. For enterprise teams this is an important signal: the company has not just pushed out another demo update, but delivered a compatible commercial upgrade without an immediate change to the price list. In Anthropic's own tests, Opus 4.

7 gains about 13% over Opus 4.6 on its internal benchmark of 93 programming tasks, including four tasks that Opus 4.6 and Sonnet 4.

6 previously failed to solve. But the most interesting point is not the performance numbers; it is that the company openly stresses the gap with Mythos. Anthropic states directly that Opus 4.

7 is less broadly capable and less dangerous in the cyber domain than Mythos Preview, and that during training the team specifically experimented with reducing such capabilities. On top of that, the model received automatic safeguards that are supposed to detect and block requests tied to prohibited or high-risk cybersecurity use. The company has a separate verification program for legitimate researchers.

This is an important shift for the whole market. Not long ago, AI developers competed mostly on who could show the best benchmark and get a flagship into public access faster. Now a different logic is emerging: the strongest model does not necessarily go mass-market if its usefulness to attackers grows too fast along with its quality.

In Anthropic's case the result is a two-tier scheme. Mythos remains in restricted access as a tool for secured and controlled scenarios, while Opus 4.7 becomes the public workhorse for code, automation and enterprise agents.

For developers this is good news, because they get a markedly stronger tool right now. For the industry as a whole, it is a reminder that the next AI race will be run not only over intelligence, but also over how controllably and in what measured doses companies can release their most powerful systems.