The release combines stronger task execution claims with new controls for high-risk cybersecurity requests.
What happened
Anthropic says the model works with higher-resolution images and is better at checking its output during complex tasks. It launched across Claude products, the API and several cloud platforms at unchanged base token prices.
Why it matters
Verification behavior matters when an agent operates for longer periods without close supervision. New safeguards also mean capability alone does not describe which requests a product will allow.
The bigger picture
Our take. Look for evidence that checks catch meaningful errors, rather than simply adding more steps. Anthropic's benchmarks and early-user reports support the launch narrative, but an upgrade still deserves evaluation on a team's own repositories and documents.
Source checked November 18, 2026. Company claims are attributed; paragraphs marked “Our take” are editorial interpretation.