Claude Opus 4.7's limits

Anthropic is keeping its most powerful Claude model, Mythos Preview, inside a tightly limited security program while selling Opus 4.7 as the broadly available version. (anthropic.com) Anthropic launched Claude Opus 4.7 on April 16, 2026 and said it is now generally available across Claude products, the application programming interface, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The company said Opus 4.7 improves on Opus 4.6 in advanced software engineering, instruction-following, vision, and long-running agent tasks. (anthropic.com) At the same time, Anthropic said Opus 4.7 is “less broadly capable” than Claude Mythos Preview, the unreleased model it introduced on April 7, 2026 through Project Glasswing. Anthropic said Mythos Preview is especially strong at computer security tasks and has already found thousands of high-severity vulnerabilities, including flaws in every major operating system and web browser. (anthropic.com 1) (anthropic.com 2) Project Glasswing is Anthropic’s gated rollout for that stronger model. The company named 12 launch partners — including Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, NVIDIA, and Palo Alto Networks — and said it also extended access to “over 40 additional organizations” that build or maintain critical software infrastructure. (anthropic.com) That is where the “50-company firewall” framing comes from: 12 named launch partners plus more than 40 additional organizations, all inside a defensive-security program rather than a public product launch. Anthropic has not announced public access to Mythos Preview and told CNBC it does not plan to make the preview generally available. (anthropic.com) (cnbc.com) The split between the two models is mostly about cyber risk. Anthropic said Opus 4.7 is the first lower-capability model released with automatic safeguards that detect and block prohibited or high-risk cybersecurity requests, and that the company wants to learn from that deployment before any broad release of “Mythos-class” models. (anthropic.com) Anthropic’s own technical write-up on Mythos Preview explains why the company is moving slowly. It said the model was capable, when directed by a user, of identifying and exploiting zero-day vulnerabilities in every major operating system and every major web browser during testing, and said more than 99% of the vulnerabilities it found had not yet been patched. (red.anthropic.com) Anthropic is also trying to channel legitimate security use into a separate lane. For Opus 4.7, the company said security professionals can apply to a Cyber Verification Program for uses such as vulnerability research, penetration testing, and red-teaming. (anthropic.com) The commercial positioning is clearer than the branding. Opus 4.7 is the model Anthropic is willing to sell widely for coding and enterprise workflows today, at the same $5 per million input tokens and $25 per million output tokens as Opus 4.6, while Mythos stays in a supervised pilot for a few dozen infrastructure and security organizations. (anthropic.com 1) (anthropic.com 2) So the limit here is not a hidden consumer feature cap. It is a deployment boundary: Opus 4.7 is the public model, and Mythos Preview remains behind Anthropic’s Glasswing partner wall until the company decides its cyber safeguards can hold up in wider use. (anthropic.com) (cnbc.com)