Anthropic Opus 4.7
- Anthropic released Claude Opus 4.7 with documented improvements in coding, tool use, memory, and long‑horizon agentic tasks. - Company and Cloudflare docs describe Opus 4.7 as Anthropic's most capable generally available model for memory and autonomous workflows. - The release signals evaluation shifting toward sustained workflow completion metrics like task completion, correction rate, and escalation frequency. (platform.claude.com)
Large language models predict the next token, but “agent” systems try to finish whole jobs across many steps. Anthropic says Claude Opus 4.7, released April 16, is built for that longer kind of work. (anthropic.com) Anthropic’s launch post says Opus 4.7 is generally available now and improves on Opus 4.6 in advanced software engineering, especially on harder coding tasks that previously needed closer supervision. The company’s API release notes say it kept the same $5 per million input tokens and $25 per million output tokens as Opus 4.6. (anthropic.com) (platform.claude.com) Anthropic’s model docs describe Opus 4.7 as its most capable generally available model for long-horizon agentic work, knowledge work, vision, and memory. Cloudflare’s model directory uses nearly the same language and lists the model as `anthropic/claude-opus-4.7` in Workers AI. (platform.claude.com) (developers.cloudflare.com) Memory is the practical issue underneath this release: most application calls are stateless, meaning the model answers once and forgets. Cloudflare’s Agents docs frame the gap directly, saying real agents need to keep state, call tools, and act without a user present on every step. (developers.cloudflare.com 1) (developers.cloudflare.com 2) That is why the benchmarks around models like Opus 4.7 are shifting from one-shot question answering to workflow measures such as whether the system completes a task, fixes its own mistakes, or escalates when it gets stuck. Anthropic’s own launch materials emphasize autonomy, steerability, and sustained coding performance rather than a single headline test score. (platform.claude.com) (claude.com) The release also comes with migration warnings for developers. Anthropic says Opus 4.7 has API breaking changes versus Opus 4.6 and points users to a migration guide before upgrading production systems. (platform.claude.com 1) (platform.claude.com 2) Anthropic’s usage guide says Opus 4.7 follows instructions more literally than Opus 4.6, which means prompts that were “good enough” on the older model may need tighter wording on the new one. The company also recommends Opus 4.7 first for the most complex tasks in its model overview. (claude.com) (platform.claude.com) The broader pattern is that model vendors are selling less on chatbot fluency and more on whether a system can carry work across minutes, hours, or weeks. Opus 4.7 fits that pitch: not just answering a prompt, but staying on task long enough to finish one. (developers.cloudflare.com) (platform.claude.com)
