Anthropic launched Claude Sonnet 5 on June 30 and made it the default model for Free and Pro users. It is also available to Max, Team, and Enterprise users, in Claude Code, and through the Claude Platform API as claude-sonnet-5.
The positioning is the important part. Anthropic says Sonnet 5 narrows the gap with Opus 4.8 on agentic work while staying cheaper. The introductory API price is $2 per million input tokens and $10 per million output tokens through August 31, 2026. After that, Anthropic says the price moves to $3 per million input tokens and $15 per million output tokens. Opus 4.8 is listed in the same post at $5 per million input tokens and $25 per million output tokens.
That makes Sonnet 5 a distribution story, not only a model story. Anthropic is trying to move more agentic behavior into the model many users meet first.
The default model is the product decision
Frontier model announcements often focus on the top system. This release matters because Anthropic is putting the new Sonnet in the default path for large parts of the user base.
Default status changes adoption. A user who never opens a model picker now gets a model Anthropic describes as better at planning, tool use, coding, and professional work. A developer using Claude Code can try the same family of capabilities without moving straight to the most expensive Opus tier. An enterprise admin can evaluate Sonnet 5 as the routine execution layer while reserving Opus for harder tasks.
That is the practical shape of the release. Anthropic is not saying Sonnet 5 replaces Opus 4.8. It is saying many workflows that previously required a larger model can now be tested on Sonnet first.
For builders, that changes the first question from “can this task justify Opus?” to “where does Sonnet fail clearly enough that Opus earns the upgrade?”
Price is part of the agent claim
Agentic models are not judged only by output quality. They also run loops. They call tools, inspect results, revise plans, and sometimes spend many turns on a task before the user sees the final answer.
That makes price more than a billing detail. A model that is slightly weaker but much cheaper can be more useful for routine agent work if it completes enough tasks reliably. A model that is stronger but expensive may still be the better choice for high-stakes or low-volume work.
Anthropic’s post leans into that trade-off. It says Sonnet 5 covers a wider range of cost-performance options than Sonnet 4.6, and that at higher effort levels it can match Opus 4.8 on some tasks. Those are Anthropic’s benchmark and product claims, not independent proof. But they point to the direction of the market: labs are starting to sell a portfolio of reasoning effort, model class, and price rather than a single “best” model.
That matters for software teams building agents. They need a routing policy, not a favorite model. Simple retrieval, refactoring, and support tasks may run on a default model. Hard planning, ambiguous security review, or expensive code changes may route upward.
Safety is also tiered
Anthropic says Sonnet 5 has a lower rate of undesirable behavior than Sonnet 4.6 in its safety assessments and a much lower ability to perform cybersecurity tasks than its current Opus models. That claim sits next to a broader week of Anthropic safety pressure around Fable 5 and Mythos 5, where government access controls and jailbreak classification became central.
For users, the safety read is narrower: Anthropic is presenting Sonnet 5 as the safer, cheaper, broadly available execution model. Opus remains the higher-end option. Fable and Mythos sit in a more sensitive class where model capability and cyber safeguards are under heavier scrutiny.
That split is useful if it stays legible. Buyers do not just need a model table. They need to know which model is appropriate for routine coding, which one is appropriate for deep agent work, which one triggers stricter review, and how usage credits behave across plans.
