Anthropic launches Claude Sonnet 5

Anthropic released Claude Sonnet 5, built, the company said, specifically for autonomous agent work — planning, tool use and independent multi-step task execution — rather than as a general-purpose chat upgrade. It sits in the middle of Anthropic’s three-tier lineup, above Haiku and below Opus 4.8, and replaces Sonnet 4.6 at that tier.

Anthropic’s central claim was cost-efficiency rather than a new capability ceiling: “Sonnet 5 narrows the gap: its performance is close to that of Opus 4.8, but at lower prices.” Standard pricing was set at $3 per million input tokens and $15 per million output tokens — about 60% of Opus 4.8’s $5/$25 — with an introductory rate of $2/$10 running through the end of August 2026. Anthropic said Sonnet 5 improved on its predecessor across reasoning, tool use, coding and knowledge-work evaluations, performing strongly on the BrowseComp agentic-search and OSWorld-Verified computer-use benchmarks, and that on some higher-effort tasks it could match Opus 4.8’s results outright.

Anthropic also noted that Sonnet 5’s cybersecurity capability trailed the Opus-tier models, which it framed as reducing misuse risk for the more widely and cheaply deployed mid-tier model. The release extended Anthropic’s practice through 2026 of shipping a cheaper, close-to-frontier model shortly after each new flagship, widening the range of cost-to-performance options available to developers building agentic products on Claude.