Anthropic: Claude Sonnet 5 brings agentic capability to the mid-tier

Anthropic released Claude Sonnet 5 on June 30, 2026, making it the default model for free and Pro tiers. The model can plan, use tools such as browsers and terminals, and run autonomously at performance levels previously associated with larger, more expensive models. Compared to Sonnet 4.6, it shows improved reasoning, tool use, coding, and knowledge work.

Pricing at launch (through August 31, 2026) is $2 per million input tokens and $10 per million output tokens, rising to $3/$15 afterward. This undercuts Opus 4.8 substantially while matching or exceeding its agentic behavior in most workflows.

The practical change for product teams is not just cost. Before Sonnet 5, teams building agentic features had to choose between a capable but expensive model and a cheaper one that struggled with multi-step reasoning. Sonnet 5 narrows that gap, making browser automation, code execution, and multi-step research tasks viable at lower operational cost. For teams working with tight compute budgets, the threshold for what is worth building in production just moved.

Anthropic also made safety improvements explicit: the model declines harmful requests more consistently and hallucinates less than its predecessor. For product managers shepherding AI deployments through legal and compliance reviews, these properties often matter as much as benchmark scores. A system that produces fewer unpredictable outputs is easier to test, audit, and present to risk teams.

The release also continues a pricing trend worth tracking. As midsize models approach the performance of last year’s frontier models, the cost of agentic AI drops with each generation. Teams that deferred agentic features due to cost should revisit those decisions against current pricing.