Claude Opus 4.7

Claude Opus 4.7 arrived on 16 April 2026, ten weeks after Opus 4.6, and was Anthropic's flagship until Opus 4.8 replaced it at the end of May. It is a refinement release rather than a redesign: the same one-million-token context window in beta, the same 128,000-token output ceiling, and the same price of five and twenty-five dollars per million tokens. What changed sits in the results. Anthropic reported 87.6 percent on SWE-bench Verified, up from 80.8 percent for Opus 4.6, together with 94.2 percent on GPQA Diamond, 69.4 percent on Terminal-Bench 2.0 and 64.4 percent on the Finance Agent benchmark — the last a state-of-the-art figure at the time. The company framed the gains as concentrated on the hardest tasks rather than spread evenly. Three practical additions came with it. The adaptive thinking scale gained an extra-high level, giving five settings for trading latency against reasoning depth. Task budgets entered public beta so a caller can cap how many tokens an agent spends on a job. And vision was widened to images up to 2,576 pixels on the long edge, roughly 3.75 megapixels, which matters for screenshots and documents. Claude Code gained an /ultrareview command for dedicated review sessions, and file-system-based memory was improved for work spanning several sessions. On safety, Anthropic shipped automated safeguards that detect and block high-risk cybersecurity requests, alongside a Cyber Verification Program for security professionals with a legitimate need. The model remains available as a previous-generation option under the identifier claude-opus-4-7; wujec.ai keeps this profile separate from Opus 4.6 and Opus 4.8.

▸Videos

No videos yet.

▸Reviews