Anthropic released Claude Sonnet 5 on June 30, positioning it as the company's most agentic Sonnet model. It can plan, use browsers and terminals, and continue multi-step work that previously required a larger model.
Sonnet 5 became the default model for Claude Free and Pro plans at launch and is also available through Claude Code, paid team plans, and the Claude API.
A smaller gap between model tiers
Anthropic says the model approaches the performance of its larger Opus line on some agentic tasks while operating at a lower price. The release focuses on coding, tool use, professional work, and sustained execution.
That balance matters for production agents because a small per-token difference can become a large operating cost across long workflows and repeated tool calls.
Effort becomes a product control
Users can select different effort levels to trade speed and cost for deeper work. This makes model behavior less like a fixed product tier and more like a resource that can be tuned for the importance of the task.
Anthropic launched introductory API pricing through August 31, 2026, before the model moves to its standard rate.
Better follow-through is the real benchmark
Agent benchmarks are useful, but teams experience the model through whether it completes a messy task, checks its work, and recovers from an error without losing the original goal.
Sonnet 5's practical value will depend on that reliability across real repositories and business systems. Human review, scoped tools, and verification remain essential even when the model appears more autonomous.