Claude’s Sonnet 5 is built to do more on its own and cost you less

Every major AI lab is racing to prove its models can work autonomously with minimal hand-holding; we’re now seeing pricing emerge as the next battleground. 

Anthropic just fired its latest shot, Claude Sonnet 5, a model the company says performs nearly as well as its flagship Opus 4.8 at a fraction of the cost.

Introducing Claude Sonnet 5, our most agentic Sonnet yet.

It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models. pic.twitter.com/UKK8G7ww5h

— Claude (@claudeai) June 30, 2026

So what’s actually new here?

Sonnet 5 is Anthropic’s most agentic Sonnet model yet. It can plan multi-step tasks, use tools like browsers and terminals, and complete work autonomously. Previously, doing that required a larger, more expensive model.

On one agentic coding benchmark, Sonnet 5 scores 63.2%, a meaningful jump over Sonnet 4.6’s 58.1%. However, it still trails Opus 4.8’s 69.2%. 

On knowledge-work tasks, though, Sonnet 5 slightly edges out Opus 4.8, but it is sort of given, especially since Opus is built for harder judgment calls. 

What does it cost, and is it actually safer?

Sonnet 5 launches today at $2 per million input tokens and $10 per million output tokens through August 31, 2026. After that, the prices increase to $3 and $15, respectively. 

That undercuts Opus 4.8, OpenAI’s GPT-5.5, and Google’s Gemini 3.1 Pro, though Gemini 3.5 Flash still remains cheaper. 

On the safety front, Anthropic reports Sonnet 5 hallucinates and shows sycophantic behavior less often than its predecessor. Furthermore, the AI model is notably weaker at dangerous cybersecurity tasks than Opus-class models, a deliberate tradeoff rather than an accident (via Anthropic). 

Anthropic’s Sonnet 5 is now available as the default model on the Free and Pro plans. It is accessible across Max, Team, Enterprise, Claude Code, and the API.

Sonnet 5’s launch follows a pattern set by rivals: OpenAI’s GPT-5.6 Sol entered preview just last week with subagent task-splitting, and Google’s Gemini 3.5 Flash, launched in May, is pitched explicitly as agentic rather than conversational.

Sonnet 5 also uses an updated tokenizer that can map the same input to up to 1.35x as many tokens as Sonnet 4.6, though introductory pricing is designed to offset that change.