Anthropic Launches Claude Sonnet 5 With Major Gains in Agentic Performance

Anthropic has released Claude Sonnet 5, bringing major improvements to its agentic capabilities. The new model narrows the performance gap with the company's Opus lineup while offering lower pricing.



The new model is designed to operate autonomously for longer stretches. The company says Sonnet 5 can make plans, navigate browsers, and use terminal commands at a level that previously required larger and more expensive models. Benchmark results show Sonnet 5 outperforms Sonnet 4.6 across reasoning, coding, tool use, and knowledge work.


At medium effort levels, Sonnet 5 delivers substantially improved cost efficiency over Sonnet 4.6. At higher effort settings, its performance can match Claude Opus 4.8 on certain complex tasks, allowing developers to balance cost and performance based on the needs of a project.



Safety evaluations indicate the model is generally safer in agentic contexts, with lower rates of hallucination and sycophancy than Sonnet 4.6. Anthropic also says Sonnet 5 performs substantially worse than its Opus models on dangerous cybersecurity evaluations, never producing a fully working software exploit during testing. As a result, Anthropic launched Sonnet 5 with its standard cyber safeguards enabled by default rather than the more restrictive protections that led the U.S. government to suspend the company's Fable 5 and Mythos 5 models.

Claude Sonnet 5 is available today as the default model for Free and Pro plans, with Max, Team, and Enterprise users also receiving access. It is also available through the Claude API and in Claude Code. To ease the transition, Anthropic is offering introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026. After that date, the cost will increase to $3 per million input tokens and $15 per million output tokens. The company also increased rate limits across Chat, Cowork, Claude Code, and the Claude Platform to accommodate higher-effort usage. Sonnet 5 uses an updated tokenizer, meaning the same text input may map to roughly 1.0–1.35× as many tokens as previous iterations, depending on the content type.