Anthropic Launches Claude Sonnet 5 for Agentic Workloads

On June 30, Anthropic released Claude Sonnet 5 — a new mid-tier model that, according to the company, handles agentic tasks at a level that previously required top-tier, high-cost flagships. That same day, the model became the default for all users — including the free plan and paid Pro subscriptions.

1Pricing and Benchmarks

Launch pricing (through August 31): $2 per million input tokens and $10 per million output tokens. After August, input tokens rise to $3. That puts it below OpenAI GPT-5.5 and Google Gemini 3.1 Pro, though above Gemini 3.5 Flash.

On agentic coding (SWE-bench), Sonnet 5 scores 63.2% — ahead of Sonnet 4.6's 58.1%, but behind flagship Opus 4.8's 69.2%. On knowledge work tasks, the new model edges out Opus 4.8. Otherwise: improvements across reasoning, tool use, coding, and general knowledge compared to the previous version.

2What Changed on Safety

Anthropic separately highlights a reduction in unwanted behaviors: compared to Sonnet 4.6, the model is less likely to assist with misuse, less likely to deceive, less likely to hallucinate, and less prone to sycophancy — telling users what they want to hear at the expense of accuracy. Its direct competitors are OpenAI's GPT-5.6 Sol and Google's Gemini 3.5 Flash.

The arrival of Sonnet 5 means the mid-tier price segment can now handle tasks that previously demanded the most expensive models. For teams building agentic systems, this directly cuts the cost per run — without sacrificing quality on most tasks.