On 30 June Anthropic released Claude Sonnet 5 - its most agentic Sonnet yet. It doesn't just answer a prompt. It plans, uses a browser and a terminal, and runs on its own. 1 million tokens of context, close to Opus 4.8, at $2 input / $10 output.
- Anthropic released Claude Sonnet 5 (30 June): the most agentic Sonnet, plans and runs on its own, 1 million tokens of context.
- Launch price $2/$10 per million tokens; the planned move to 3/15 was dropped - on 10 August Anthropic made it permanent.
- Agentic coding 63.2% (Opus 4.8 sits at 69.2%); already the default in Claude Code and for Free/Pro on claude.ai.
Anthropic released Claude Sonnet 5 on 30 June 2026 - its most agentic Sonnet yet. It doesn't just answer a prompt. It plans steps, calls tools, reads the result, and keeps going on its own. Without anyone nudging it at every step. It's already the default choice in Claude Code and for free and Pro users on claude.ai.
Updated 1 October 2026: on 10 August Anthropic announced that the launch price of $2 input and $10 output per million tokens is now permanent. The planned move to 3/15 from 1 September no longer applies.
In practice the news isn't 'another model.' The news is that agentic work got cheaper. Until now, an agent that runs on its own for hours came out expensive. At this price it goes mass-market. That simple. The cheaper agent that's 'good enough' and doesn't stop does more real work than the pricier genius you have to watch at every step. I tried it myself on a project - decent, not top-tier, but at this price it'll get the job done.