Future TechnologyClaude Sonnet 5 is here, and it might be the most useful AI Anthropic has shipped
Anthropic launched Claude Sonnet 5 on June 30, and it's a genuine step up from what Sonnet 4.6 could do. The headline number is the price: $2 per million input tokens, $10 per million output, as an introductory rate through August 31, after which it moves to $3/$15. That's still cheaper than the previous Sonnet at launch, for a model that Anthropic says performs close to Opus 4.8.
What changed under the hood matters more than the price. Sonnet 5 is built to run autonomously, which in practice means it can plan multi-step tasks, use tools like web browsers and terminals, and keep going without stopping to ask for permission at every turn. Six months ago you'd have needed a larger and more expensive model to do this reliably. Now it's the mid-tier default.
The context window
Sonnet 5 ships with a 1 million token context window by default, and supports 128,000 max output tokens. For anyone building agents that need to hold long documents in memory, process entire codebases, or maintain state across a complex task, that's a meaningful upgrade. The 1M context was previously a premium feature tied to larger models.
Who it's for
Anthropic has made Sonnet 5 the default model for Free and Pro Claude plans, so most users get it without doing anything. For developers, it's available as claude-sonnet-5 via the API, through Amazon Bedrock, and on Google Cloud's Vertex AI.
The standout use case is agentic work: running Claude as a background worker that completes tasks end-to-end rather than answering one question at a time. Sonnet 5's improvements in tool use, reasoning, and software coding are specifically tuned for that. If you've been using Claude Code, you'll notice the difference quickly.
What it doesn't do
Sonnet 5 isn't Opus 4.8. For the most demanding reasoning tasks, complex research synthesis, or work where you need the absolute ceiling, the larger model still has an edge. The introductory pricing also runs out at the end of August, so worth building cost assumptions around the standard $3/$15 rate rather than the promo.
On balance, Sonnet 5 is the model most developers should be defaulting to right now. It does Opus-level things at Sonnet prices, at least for a few months.