Anthropic released Claude Sonnet 5 on June 30, 2026, bringing autonomous agent capabilities to its mid-tier model line that until recently required the company’s larger and costlier Opus-class models.
What changed
Anthropic describes Sonnet 5 as its “most agentic Sonnet yet.” The model can draw up multi-step plans, use external tools including web browsers and computer terminals, and run long tasks without human check-ins — performance the company says now approaches Opus 4.8 while costing significantly less.
Benchmark scores back up the claim: 46.8% on Humanity’s Last Exam (with tools) and 78.5% on OSWorld-Verified, a computer-use task completion test. Anthropic says reasoning, coding, and knowledge-work scores are all substantially higher than in Sonnet 4.6.
Pricing and availability
Sonnet 5 is immediately available across all Claude plans — Free, Pro, Max, Team, and Enterprise — and via the API under the identifier claude-sonnet-5. Through August 31, 2026, the introductory API rate is $2 per million input tokens and $10 per million output tokens; standard pricing of $3/$15 per million tokens takes effect on September 1.
Safety
Anthropic highlighted safety alongside capability gains. Sonnet 5 has a lower rate of undesirable behaviors than Sonnet 4.6, is better at rejecting malicious prompts and resisting prompt injection, and ships with cyber safeguards enabled by default — protections previously deployed only on Opus models.