Anthropic releases Claude Opus 4.7 for agentic programming, vision, and autonomous tasks

Anthropic released Claude Opus 4.7 on April 16, 2026 — a targeted yet highly significant update to its flagship model that addresses practical developer pain points rather than abstract benchmarks: agentic programming, handling detailed images, and long autonomous tasks where the model must not just answer, but plan, verify itself, and complete tasks end-to-end. The company positions Opus 4.

7 as a direct upgrade to Opus 4.6, not a new product line. Yet in engineering scenarios, the difference proved substantial.

According to Anthropic, the model better follows instructions, handles multi-step processes more carefully, and more often finds ways to verify results before returning them to users. This is particularly important in environments where AI does not simply write code snippets on demand, but acts as a semi-autonomous executor: reading repositories, running tools, fixing errors, continuing work after failures, and maintaining intermediate solutions in memory. On Anthropic and partner benchmarks, the improvement appears far from cosmetic.

On Anthropic's internal 93-task programming benchmark, Opus 4.7 improved the task completion rate by 13% compared to Opus 4.6, including four tasks that neither Opus 4.

6 nor Sonnet 4.6 could solve. In CursorBench, the model achieved 70% versus 58% in the previous version.

In Notion's test cases for complex multi-step processes, the improvement was 14% with lower token consumption and three times fewer tool-calling errors. Rakuten separately states that on Rakuten-SWE-Bench, the new model closes three times more production tasks than Opus 4.6.

The second major upgrade is vision. Claude Opus 4.7 can process images up to 2576 pixels along the longer side—approximately 3.

75 megapixels. This is more than three times higher than the limit of previous Claude models. For normal chat mode, such a figure might seem secondary, but for agentic scenarios it is critical: the model can parse dense interface screenshots, technical diagrams, schematics, chemical structures, and documents where fine details matter.

Anthropic provides an example from computer use: on the XBOW visual benchmark, the new version scored 98.5% versus 54.5% for Opus 4.

6. In effect, this moves a range of tasks from the category "sometimes works" to "can be integrated into products." Anthropic has also strengthened tools for long autonomous execution.

Opus 4.7 introduces a new reasoning effort level — xhigh, positioned between high and max. In the API, the company launched a public beta of task budgets, allowing developers to cap the token budget on long runs and manage step priorities.

Claude Code now features an ultrareview mode for deeper examination of changes, and auto mode now allows fewer interruptions for long tasks with permission requests. However, migration from Opus 4.6 is not entirely free: the updated tokenizer can increase input tokens by approximately 1.

0–1.35x depending on content type, and deeper reasoning in later stages of agentic tasks increases output token volume. Anthropic separately emphasizes the security theme.

Opus 4.7 launches shortly after the announcement of Mythos Preview — a more powerful model whose access Anthropic decided to restrict due to cybersecurity risks. Therefore, Opus 4.

7 is the first public model where Anthropic tests automatic blocking for high-risk cyber requests. For legitimate scenarios such as pentesting, red teaming, and vulnerability research, the company simultaneously opened a verification program. By Anthropic's own assessment, the security profile of Opus 4.

7 is generally close to 4.6: the model improved in honesty and resistance to prompt injection, though in certain categories it is imperfect and still trails Mythos Preview in overall behavior consistency. The main conclusion is simple: Claude Opus 4.

7 is not a new generation for a new number's sake, but a pragmatic release for those building real workflows on top of the model. If your scenario is an IDE assistant, a code review agent, document analysis, computer vision for interfaces, or long chains of tool-driven actions, the update looks significant right now. For the average user, the difference may not always be obvious, but for developers and teams that measure quality not by demos but by the number of tasks completed end-to-end, Opus 4.

7 looks like one of Anthropic's most useful updates in recent times.