Anthropic launched Claude Sonnet 5 on June 30, and it’s the summer’s biggest model news for one very concrete reason: what required the largest, priciest models a few months ago — agents that plan, use a browser and a terminal, and work on their own — is now in the mid-size model, at mid-size prices. Here’s what it brings, what it costs, and our honest take.
What Sonnet 5 is
- The most agentic Sonnet to date: it plans multi-step tasks and uses tools (browser, terminal) autonomously, according to Anthropic’s official announcement.
- Performance close to Opus 4.8: on agentic coding it scores 63.2% versus Opus 4.8’s 69.2% and the previous Sonnet’s 58.1%. Not the flagship, but remarkably close for far less money.
- A 1-million-token context window by default, up to 128k output tokens and adaptive thinking (it reasons longer only when the task demands it).
- It’s the new default model on Free and Pro plans: most Claude users already have it without touching anything.
The price is the whole story
On the API it launches at an introductory $2 per million input tokens and $10 output until August 31, 2026; afterwards $3/$15. To put it in perspective: near-flagship performance at a fraction of flagship cost. If you have something to test with agents, these discounted weeks are the moment.
Our take: what changes for you
- If you use Claude Free or Pro: you already have it. What you’ll notice most is longer tasks holding together: multi-step instructions without losing the thread halfway.
- If you build on the API: this is the year’s big move in the agent price war. Jobs you budgeted for the large model now cost half; redo your maths before your competitors do.
- The safety nuance almost nobody will report: it’s the first Sonnet-tier model with real-time cybersecurity safeguards — high-risk requests in that domain may be refused. For 99% of uses you won’t notice, and as a trend we think it’s the right signal: cheap autonomous agents need brakes fitted as standard.
Our honest reading: the story isn’t the benchmark, it’s the democratisation. When agentic capability drops into the free plan’s default model, it stops being a lab demo and starts being an everyday tool.
Frequently asked questions
Does it replace Opus 4.8?
No: Opus stays ahead on the hardest tasks. Sonnet 5 is the quality-price sweet spot; Opus is the specialist for the difficult stuff.
Do I need to do anything to use it?
On Free and Pro, nothing: it’s already the default. On the API, just point to the new model — and catch the introductory pricing before August 31.
Conclusion
Sonnet 5 pushes the price of competent agents down, and that moves the whole board. For family context, see Claude Opus 4.8 and Claude Fable 5; and to understand what “agentic” really means, the state of AI agents.
