What Is Claude Sonnet 5? Near-Opus Performance and Pricing Explained

What Is Claude Sonnet 5?

Claude Sonnet 5 is the latest model in the Sonnet series of Claude, released by Anthropic on June 30, 2026. Sonnet was originally positioned as a mid-tier model with a good balance of performance and price. This Sonnet 5, however, goes beyond that frame: its biggest feature is offering performance that approaches the flagship Opus 4.8 while keeping the price down. Let's look at what it is and its features in order.

The Most Agentic Sonnet

Claude Sonnet 5 is a model with enhanced ability as an "agent" that plans on its own and works autonomously using tools such as browsers and terminals. By agent here, we mean a use case where it takes an instruction just once, decides the intermediate steps on its own, and sees the task through to the end. Anthropic describes Sonnet 5 as the most agentic Sonnet yet.

Until recently, this kind of autonomous work required much larger and more expensive models. The significance of Sonnet 5 is that it delivers work at that level in a mid-tier price range.

Near-Opus 4.8 Performance at a Low Price

Another feature of Sonnet 5 is how close it is to the flagship model. Anthropic explains that Sonnet 5's performance has narrowed to a level close to Opus 4.8, and yet its price is lower.

In other words, the key point this time is that much of the work that previously required a flagship model can now be handled by the more affordable Sonnet 5. We'll confirm the specific differences with numbers in the benchmarks section that follows.

1M-Token Support

Sonnet 5 can also handle a large amount of text at once, supporting 1 million tokens on the API and major cloud platforms. A token is a small unit that text is broken into, and 1 million tokens is roughly equivalent to one or two books.

It is also equipped with a mechanism to keep track of its remaining capacity (its remaining tokens) as it proceeds through a long exchange, making it well suited to loading lengthy materials all at once or being entrusted with long-running work.

Sonnet 5's Performance and Benchmarks

Anthropic presents benchmark results across multiple areas, including coding, reasoning, and PC operation. In each benchmark, the bars are listed from top to bottom in the order Sonnet 5, the previous-generation Sonnet 4.6, and the flagship Opus 4.8 (reference value). In every case you can see that it has clearly improved over Sonnet 4.6 and has come very close to Opus 4.8.

Agentic Coding Performance

Sonnet 5 has greatly improved its scores from the previous generation on benchmarks that simulate real software development. On SWE-bench Pro, which evaluates difficult code fixes, it scores 63.2%, beating Sonnet 4.6's 58.1%. On Terminal-Bench 2.1, which measures command-line work that requires planning, iteration, and tool integration, it scores 80.4%, a big step forward from Sonnet 4.6's 67.0%.

Scores in Reasoning, PC Operation, and Knowledge Work

It has also advanced beyond coding. On Humanity's Last Exam, which gathers difficult cross-domain questions; OSWorld-Verified, which operates a PC screen; and GDPval-AA v2, which measures practical knowledge work, it surpassed Sonnet 4.6. The table below summarizes the main scores presented by Anthropic.

GDPval-AA v2, which measures knowledge work, is shown as a score rather than a percentage, and Sonnet 5's 1618 slightly exceeds Opus 4.8's 1615. Depending on the area, there are now cases where Sonnet 5 stands shoulder to shoulder with, or even overtakes, the flagship model.

Improvement from Sonnet 4.6 and the Gap with Opus

Lining up the numbers makes the trend clear. Compared with the previous-generation Sonnet 4.6, Sonnet 5 has improved on every item by anywhere from a few points to more than ten points. On the other hand, the gap with the flagship Opus 4.8 stays within a few points on many items. Sonnet 5's real strength is that it has been reliably raised up from the old Sonnet and has come to just short of the flagship. How you view this gap becomes the basis for choosing a model in the second half.

Sonnet 5's Pricing and Available Plans

Pricing is easier to grasp if you think of it in two ways of using it: chat under a monthly plan, and the API where you pay for what you use. Let's look at each in turn.

What You Can Use in Chat, and the Default Model

First, chat, which most people use. Claude Sonnet 5 has become the default model on the free plan (Free) and the paid Pro plan. That means you can converse with Sonnet 5 as standard, with no special setup. It is also available on the Max, Team, and Enterprise plans, and within a monthly plan, there is no per-model additional charge for using chat.

Now that Sonnet 5 has become the standard for Free and Pro, many users are naturally using a higher-performing model than before. For how to use Claude overall and how to think about plans, see also the explainer article What Is Claude? Pricing and How to Choose a Model.

API Pricing and the Introductory Price

The API, which you build into your own app or program, is usage-based according to the amount used (tokens). Sonnet 5 has an introductory price set: through August 31, 2026, it is $2 per million input tokens and $10 per million output tokens, and after that it becomes $3 for input and $15 for output.

For the time being, it's an introductory-price period during which you can use it at a relative bargain for its performance. For how to compare pricing with the Opus series, we lay it out in detail in the explainer article Claude Opus 4.8 / 4.7 / 4.6 Pricing Comparison.

Availability in Claude Code and the API

For developers, it is available from Claude Code, the AI coding environment, and the Claude API. The model ID (the model's identifier) to specify in the API is claude-sonnet-5. In addition, it can be used from major cloud platforms such as Amazon Bedrock, Google Cloud, and Microsoft Foundry.

How to Choose Between Sonnet 5 and Opus

With the arrival of Sonnet 5, choosing between it and the flagship Opus 4.8 has become clearer. The benchmark gap is small and Sonnet 5 is cheaper, so the axis of judgment narrows down to "how much accuracy do you need."

When Sonnet 5 Is a Good Fit

Most everyday work can be handled just fine with Sonnet 5. It broadly suits uses where the balance of performance and cost is effective, such as writing and summarizing text, research, programming assistance, and loading long materials. In particular, for work that runs many similar processes or situations where you want to keep costs down, Sonnet 5 — which offers near-Opus 4.8 performance cheaply — is the first choice.

When You Should Choose Opus 4.8

On the other hand, for work where the last slight bit of accuracy greatly affects the outcome, Opus 4.8 still has the edge. For example, on SWE-bench Pro, Opus 4.8 scores 69.2%, beating Sonnet 5's 63.2%. For the most difficult code fixes, or analysis where you want to prioritize quality above all, Opus 4.8 is worth choosing. The features of Opus 4.8 itself are explained in the article Claude Opus 4.8 Release Explained.

The Bottom Line on Choosing a Model When Unsure

To sum up, the sensible choice, both in terms of performance and price, is to first try the standard Sonnet 5 and switch to Opus 4.8 only for the tough spots where accuracy is simply not enough. Because the Free and Pro default has become Sonnet 5, most users can enjoy this benefit without changing any settings in particular. If you want to compare with other companies' latest models, see also the article GPT-5.6 Sol, Terra, and Luna Explained.

Claude's official announcements are often long texts in English, and there are times when you want to quickly grasp the key points. If you tidy up a web page into markdown format as-is, the structure of headings and tables is preserved, and it also tends to improve the accuracy when you have an AI read and organize it.

Free ToolURL to Markdown ConverterConvert any public web page URL to Markdown. Preserves headings, tables, lists, and links — perfect for LLM and RAG preprocessing, research notes, and archiving web articles.Try it now →