What happened
On June 30, Anthropic released Claude Sonnet 5, the newest version of its mid-tier model. The company describes it as its most agentic Sonnet yet: a model that can make plans, use tools such as web browsers and terminals, and work through multi-step tasks with less supervision than earlier versions.
The headline claim is about value rather than raw records. According to Anthropic, Sonnet 5's performance approaches that of Opus 4.8, the company's flagship, while costing significantly less to run. Compared with its predecessor, Sonnet 4.6, the new model shows improvements in reasoning, tool use, coding, and general knowledge work.
Sonnet 5 is now the default model for users on Claude's Free and Pro plans, and it is available to Max, Team, and Enterprise customers. Developers can access it through the Claude API as well as AWS Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry, where it is in preview.
Pricing and positioning
Anthropic set an introductory price of 2 dollars per million input tokens and 10 dollars per million output tokens through August 31, 2026, after which standard pricing of 3 dollars and 15 dollars applies. Technology press coverage, including TechCrunch, framed the launch squarely as a cheaper way to run AI agents — the autonomous, tool-using workloads that consume far more tokens than ordinary chat.
That framing reflects where the industry's costs actually sit. An agent that browses, writes code, and checks its own work can burn through millions of tokens on a single task. For those workloads, a mid-priced model that performs close to a flagship changes the economics considerably.
The safety picture
Anthropic's release notes report lower rates of misaligned behavior than Sonnet 4.6, along with better performance at refusing malicious requests and resisting prompt-injection attacks, where hidden instructions in a web page or document try to hijack what the model does. The company also notes that Sonnet 5 has substantially weaker cybersecurity capabilities than its Opus models, and it ships with cyber safeguards enabled by default.
The timing is notable: the release came the same week the U.S. government lifted export controls on Anthropic's top-tier Fable 5 and Mythos 5 models, a reminder that the capability differences between model tiers are increasingly a security consideration as well as a pricing one.
Why it matters
- Frontier performance keeps getting cheaper. Each generation, the mid-tier model catches up to the previous flagship. Capabilities that cost premium prices a year ago are now the default for free users.
- Agents are the battleground. The major labs are competing less on chatbot quality and more on how well models can complete real, multi-step work autonomously — and on how affordable that is at scale.
- Defaults shape experience. Because Sonnet 5 is now what Free and Pro users get automatically, this release changes the day-to-day AI experience for a very large number of people who will never read a benchmark table.
What to watch next
Watch how the introductory pricing period plays out — whether competitors respond with cuts of their own — and how quickly developers shift agent workloads from flagship models down to Sonnet 5 in practice. Independent evaluations over the coming weeks will show whether the near-Opus performance claims hold up outside Anthropic's own testing. And with OpenAI's GPT-5.6 family in limited preview at the same time, the next round of the model race is already underway.
