Anthropic launches Claude Sonnet 5 with stronger agents and lower prices

Anthropic launched Claude Sonnet 5 with stronger agentic capabilities and lower prices, saying the model can make plans, use browsers and terminals, and run autonomously at a level that only larger, pricier systems handled a few months ago. The company said Sonnet 5 is the most agentic Sonnet model yet and that its performance comes close to Opus 4.8, while undercutting Anthropic’s own top-tier line.

The pricing is the sharpest part of the move. Anthropic set introductory rates at $2 per million input tokens and $10 per million output tokens through August 31, 2026, then said the price will rise to $3 per million input tokens and $15 per million output tokens after that date. The model is available across Free, Pro, Max, Team and Enterprise plans, and it is also accessible in Claude Code, on the Claude Platform and through the Claude API under the name claude-sonnet-5.

Anthropic is positioning the release as a continuation of the Sonnet line’s climb into more capable agent work. Sonnet 4.5 arrived on September 29, 2025 and was billed as the best model in the world for agents, coding and computer use at the same $3 and $15 pricing. Sonnet 4.6 followed on February 17, 2026 with a 1 million token context window in beta and kept the same price. Opus 4.8 launched on May 28, 2026 as Anthropic’s upgraded Opus-class model for coding, agentic tasks and long-running work.

The company is also leaning hard on safety. In its system card for Sonnet 5, Anthropic said the model showed a lower rate of undesirable behaviors than Sonnet 4.6 and was generally safer in agentic contexts. At the same time, it said Sonnet 5 had much lower ability to carry out cybersecurity tasks than current Opus models, a signal that Anthropic is drawing a line between broad autonomy and sensitive misuse. That tradeoff will matter as more businesses test whether a cheaper model can take on real work without pushing risk too far.

Sonnet 5 also fits a broader market contest that has been defined by cost as much as capability. Anthropic said recent Sonnet models, including 3.5, 3.6 and 3.7, were the first to show strong coding and tool-use skills, while the latest gains in agentic performance had mostly come from the Opus line. By narrowing the gap with Opus 4.8 and lowering the entry price, Anthropic is trying to make Sonnet the model developers reach for first, not the compromise they accept last.