Claude Sonnet 5 is Anthropic’s latest Sonnet-class model, released on June 30, 2026. Its biggest change is not just stronger answers. Anthropic has pushed Sonnet further toward agentic work, where the model needs to plan a task, use tools, make decisions, check its progress, and continue until the job is complete.

That makes it more interesting for developers, researchers, analysts, writers, and teams doing multi-step work than for basic question answering. Sonnet 5 also accepts text and image inputs, and Anthropic positions it for coding, browser and computer use, professional workflows, content work, and autonomous agents.

This is the main reason to pay attention to Sonnet 5. It can plan, work with browsers and terminals, use tools, recover from errors, and stay focused across longer workflows. Anthropic says the model closes much of the capability gap between earlier Sonnet models and its higher-end Opus line on agentic tasks.

The practical difference is follow-through. A useful test is not, “Can Claude write this function?” It is, “Can Claude inspect the project, find the cause of the bug, change the correct files, test the result, and explain what it changed?”

Sonnet has long been closely associated with coding, and Sonnet 5 continues in that direction. It is designed for planning implementations, debugging, maintaining existing projects, handling multi-file codebases, and carrying out larger refactors.

It should be particularly useful when the codebase is messy enough that the model has to investigate before writing code. That is a more meaningful test than asking it to generate a clean standalone script from scratch.

Claude’s current models can track progress through extended tasks, and Sonnet 5 includes context awareness that helps it manage how much working context remains. In supported environments, this matters for jobs that stretch across many steps rather than one large prompt and one response.

A 1-million-token context window is also supported for Sonnet 5 on Amazon Bedrock, which makes large document collections and substantial codebases more practical to work with in one workflow.

The best prompts give Sonnet a job, not just a question.

Claude responds better when the goal, constraints, available tools, expected output, and definition of success are explicit.

Sonnet 5 makes the most sense for software development, AI agents, research, document analysis, technical writing, business analysis, browser-based workflows, and multi-step knowledge work. Anthropic also makes it available through Claude.ai, Claude Code, its API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry, so it can move from individual chat use into production systems.

Sonnet 5 is more autonomous, but autonomy does not remove the need for review. Research can still contain incorrect conclusions, code should still be tested, and important business decisions need human verification.

Its built-in knowledge cutoff is January 2026, so newer facts need access to current sources or tools.

There is also a temptation to use an agentic model for every task. For a short rewrite or basic question, the extra planning ability may add little. Sonnet 5 becomes more valuable as the task gains steps, tools, files, constraints, or opportunities for the model to correct itself.