Anthropic released Claude Sonnet 5 on June 30, positioning its mid-tier model as a cost-effective workhorse for autonomous AI agents. The model matches or exceeds its predecessor Sonnet 4.6 across coding, reasoning, and tool-use benchmarks while undercutting rival offerings from OpenAI and Google.
Agentic performance at a discount
Sonnet 5 scores 63.2 percent on Anthropic’s agentic coding benchmark, up from Sonnet 4.6’s 58.1 percent and closing in on Opus 4.8’s 69.2 percent. On knowledge work benchmarks the model slightly outperforms the more expensive Opus 4.8, which was designed for deep research and nuanced judgment.
“Between Sonnet 5 and Opus 4.8, users can adjust the effort level to find the right balance of cost and performance,” Anthropic said in a blog post.
The model excels at completing complex multi-step tasks without stalling and checks its own output unprompted, capabilities that, until recently, required larger and more expensive models.
Pricing
Anthropic is offering an introductory rate of US$2 (approximately £1.60) per million input tokens and US$10 (approximately £8) per million output tokens until August 31, 2026. After that, pricing rises to US$3 (approximately £2.40) input and US$15 (approximately £12) per million output tokens.
At the promotional rate, Sonnet 5 undercuts Opus 4.8, OpenAI’s GPT-5.5, and Google’s Gemini 3.1 Pro, while remaining pricier than Google’s budget offering Gemini 3.5 Flash.
Safety improvements
The new model shows lower rates of undesirable behaviors than Sonnet 4.6, better at refusing malicious requests, more resistant to prompt-injection hijacking, and reduced hallucination and sycophancy. However, safety scores are not on par with the top-tier Opus 4.8 or Claude Mythos Preview, particularly for dangerous cybersecurity tasks.
“At Lovable, we’re putting powerful tools in the hands of millions of builders,” said Fabian Hedin, co-founder of Lovable, an AI app-building platform. “A model that knows when to say no is just as important as one that knows how to build.”
Reactions from early testers
Daniel Shepard, senior engineer at Zapier, tested Sonnet 5 on a two-part automation: updating Salesforce account tiers and sending a launch announcement to enterprise contacts. “It finished end to end. That used to stall halfway. For day-to-day automation, it’s a no-brainer,” he said.
Availability
Sonnet 5 became the default model for free and Pro tier users on June 30, and is available across all subscription tiers.
The launch underscores a broader market shift: agentic capability is now the baseline expectation at every price tier. The competitive differentiator has moved to cost and reliability without human oversight.