Anthropic pushed Claude Sonnet 5 out on June 30, 2026, and it's already the default model on free and Pro plans. If you've been putting off checking what changed since Sonnet 4.6, here's the breakdown, not the marketing version.
What's different from Sonnet 4.6
The headline change is agentic performance. Sonnet 5 can plan a task, reach for tools like a browser or terminal, and keep working without you checking in every few minutes. Anthropic says it gets close to Opus 4.8 level output on coding and reasoning tasks, at a fraction of the cost.
Two features matter more than the benchmark scores:
- A 1M-token context window, a major jump from Sonnet 4.6, so it can hold an entire codebase or a long research document without losing track of earlier context.
- Adjustable reasoning effort. You can now pick low, medium, high, max, or x-high depending on whether you want a fast answer or want the model to actually work through the problem. Low-effort mode is close to instant; x-high burns more tokens but catches edge cases a fast pass would miss.
It also takes image and file inputs alongside text, which carries over from Sonnet 4.6 but is worth confirming if you're coming from an older model.
Pricing: what it costs right now
This is the part that hits your bill.
| Through August 31, 2026 | After August 31, 2026 | |
|---|---|---|
| Input tokens | $2 per million | $3 per million |
| Output tokens | $10 per million | $15 per million |
That's a 50% price increase baked in from day one, which is a strange way to launch a model Anthropic is positioning as the cheaper option. If you're building something that leans on Sonnet 5 for agent work, plan around the higher rate now instead of getting surprised in September.
Should you switch from Sonnet 4.6 or Opus
If you're running long, multi-step coding tasks or anything that needs the model to use tools on its own, Sonnet 5 is worth the switch. That's where the real gains show up. If your use case is short Q&A or straightforward writing, you probably won't notice much difference from 4.6, and there's no rush to move.
Opus 4.8 is still the model to reach for when you need the best possible reasoning quality and cost isn't the deciding factor. Sonnet 5 is Anthropic's attempt at Opus-adjacent performance without Opus pricing, and for agentic and coding work, it mostly delivers on that.
One thing to check before you rely on it for security work
Sonnet 5 is the first Sonnet-tier model with real-time cybersecurity safeguards built in. Ask it something that resembles a prohibited or high-risk security topic and it may refuse outright, even when the use case is legitimate, like penetration testing or CTF practice. Test this before you build a workflow around it, because a refusal mid-task is a bad surprise.
The takeaway
Claude Sonnet 5 is a genuine upgrade for coding and agent work, and the 1M-token context window plus adjustable reasoning effort are the two features that matter most day to day. The pricing is good through August and gets less good right after. If you want to try it without committing to a paid plan, our guide on [how to use Claude AI for free](https://www.questloops.com/blog/use-claude-ai-free) covers how to get the most from the free tier, and you can see how Sonnet 5 stacks up against the competition in our [ChatGPT Plus vs Claude vs Perplexity comparison](https://www.questloops.com/blog/chatgpt-plus-vs-claude-vs-perplexity).
