# Anthropic's New AI model, Claude Sonnet 5, Feels Like Opus on Budget

*Published:* 2026-07-01
*Author:* Farzan Hussain

![Claude Sonnet 5 LLM Model running on Android](https://bestforandroid.com/wp-content/uploads/2026/07/Claude-Sonnet-5-LLM-Model-running-on-Android.jpg)

Anthropic just shipped a new AI model that’s supposed to make its own flagship model look overpriced. That’s a strange thing for a company to admit about its own product, and it’s exactly what caught my attention.

The model is called [Claude Sonnet 5](https://www.anthropic.com/news/claude-sonnet-5), and it launched this week as a mid-tier option that sits below Anthropic’s top-of-the-line Opus 4.8. Anthropic calls it its most agentic Sonnet model yet, with performance close to Opus 4.8 and a real improvement over its predecessor, Sonnet 4.6.

This means it can plan out multi-step tasks, use tools like browsers and terminals, and keep working without you babysitting every move.

What Claude Sonnet 5 actually does
----------------------------------

This isn’t just a chatbot upgrade.

Anthropic says it can make plans, use tools like browsers and terminals, and run autonomously at a level that used to require bigger, pricier models. If you’ve used AI coding assistants that stall out halfway through a task, that’s the exact problem this new LLM is aimed at.

One of the engineers described handing it a two-part job, updating Salesforce account tiers and sending a launch announcement to enterprise contacts, and having it finish the whole thing without stopping.

That’s the kind of task that used to need human intervention to complete the process.

Does Sonnet 5 actually beat Opus?
---------------------------------

![Claude Sonnet 5 comparison with Sonnet 4.6 and Opus 4.8](https://bestforandroid.com/wp-content/uploads/2026/07/Claude-Sonnet-5-comparison-with-Sonnet-4-6-and-Opus-4-8.jpg)

Not quite, and Anthropic isn’t pretending otherwise. On one agentic coding benchmark, Sonnet 5 scored 63.2%, compared to Opus 4.8’s 69.2% and Sonnet 4.6’s 58.1%. It closed most of the gap without closing all of it.

At its highest reasoning setting, Sonnet 5 performs roughly in line with Opus 4.8’s medium to high setting on a couple of specific benchmarks, but running it at that level costs more than just using Opus at a comparable setting.

So the “just use Sonnet 5 for everything” pitch has a limit. Moreover, Opus still wins when accuracy matters more than the bill.

The price is the actual headline
--------------------------------

Here’s where it gets interesting for anyone building with this instead of just chatting with it.

Sonnet 5 launches at $2 per million input tokens and $10 per million output tokens through August 31, before jumping to $3 and $15. That’s still less than what Opus costs, and it undercuts some of [OpenAI](https://en.wikipedia.org/wiki/OpenAI)‘s and Google’s competing mid-tier models too.

Starting this week, Sonnet 5 also becomes the default model for free and Pro plan users in the Claude app. So most people using Claude casually are already on it, whether they noticed or not.

I’ve spent enough time around AI tools and LLM models to be skeptical of “near flagship performance for less” claims. Usually, there’s a catch buried in the fine print.

However, here, the catch is honest: it’s genuinely close to Opus on a lot of tasks, genuinely behind on the hardest ones, and priced low enough that the tradeoff makes sense for most real work.

If you’re not deep into AI coding tools, this might feel like inside baseball. But it’s a preview of where every AI company is headed next, i.e., not a smarter chatbot, but a cheaper agent you can actually trust to get the job done.