Anthropic's Latest Opus Release Raises Questions About Its AI Strategy

Anthropic has launched Claude Opus 4.7, its most capable model for general public use. The company positions it as an upgrade for complex software engineering and document creation, with improved image analysis. However, its debut is shadowed by the superior performance of a different, restricted model called Mythos Preview.

Internal evaluations show Opus 4.7 underperforms compared to the cybersecurity-focused Mythos Preview on every relevant benchmark. Anthropic openly states this new public model does not push its technical frontier. Instead, Opus 4.7 serves as a controlled testbed. The company has intentionally limited its cyber capabilities and loaded it with new safeguards, using this rollout to gather data for a future, broader release of more powerful Mythos-class systems.

For now, Mythos Preview remains in private hands, accessible only to partners like Nvidia, JPMorgan Chase, and Microsoft. This tiered release strategy highlights a cautious approach to deploying advanced AI. Security researchers can apply for relaxed safeguards through Anthropic's new Cyber Verification Program.

Priced identically to its predecessor at $5 per million input tokens and $25 per million output tokens, Opus 4.7 is already in use by early customers including Intuit, Databricks, and Shopify. The launch underscores a deliberate pacing in the AI race, where the most powerful tools are being held back for further testing.

Source: The Verge