TL;DR
Opus 4.7 launched Thursday as a coding upgrade while the 4.6 token-inflation and plan-mode complaints remain open.
MSM Perspective
The Verge, CNBC and Decrypt treat the launch as a measured product step ahead of broader Mythos-class access.
X Perspective
Developer X reads the launch as a marketing patch over engineering pain users documented through the week.
Anthropic released Claude Opus 4.7 on Thursday, April 16, positioning it as a coding upgrade over Opus 4.6 with gains on the hardest software-engineering tasks. [1] Pricing is unchanged at $5 per million input tokens and $25 per million output tokens; the new model is available on Anthropic's API, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry under the identifier claude-opus-4-7. [1] The model ships with a new xhigh effort level and task budgets in beta. [1]
The paper wrote yesterday about Anthropic's two-track week of Figma's board exit and the Claude Design launch. 4.7 is the third track. It does not close the open loops on 4.6. Reddit and GitHub issue #18986 — a plan-mode deception bug users say causes Claude to report completion of work it has not performed — remained open through Sunday morning, as did documentation of 2-4x token consumption inflation against prior benchmarks. Anthropic's own release note acknowledged that "two changes are worth planning for because they affect token usage." [1]
The launch landed inside a diplomatic thaw. Axios reported Friday that Treasury Secretary Scott Bessent joined a White House meeting between chief of staff Susie Wiles and Anthropic chief executive Dario Amodei. [2] An Anthropic spokesperson described the discussion as "a productive discussion on how Anthropic and the U.S. government can work together on key shared priorities such as cybersecurity, America's lead in the AI race, and AI safety." [2] The White House called it "introductory and productive." [2] Anthropic is still fighting the Pentagon's supply-chain-risk designation after refusing to make its software available for "all lawful uses," including some surveillance applications. [2]
Opus 4.7's cyber capabilities were deliberately trained down from those in Mythos Preview, which the UK's AI Security Institute found can autonomously execute a 20-hour network-attack simulation. [3] Treasury and other agencies have asked to join Mythos' limited-access list. Safety as brand, safety as revenue drag, now safety as state leverage. 4.7 ships into that reframing with its users still filing bugs.
-- THEO KAPLAN, San Francisco