What Is Claude Sonnet 5? A Complete Overview

Claude Sonnet 5 is Anthropic's mid-tier large language model, built to plan multi-step work, use tools like browsers and terminals, and run autonomously on long tasks. Anthropic released it on June 30, 2026 and calls it the most agentic Sonnet it has shipped. The reason it matters: Sonnet 5 narrows the capability gap to Anthropic's flagship Opus line while staying at Sonnet-tier prices, which changes the calculation for anyone who defaulted to a larger, pricier model for serious agentic work.

If you have been running a heavier model because earlier Sonnets could not keep up, that assumption is worth revisiting. This overview covers what Sonnet 5 is, what changed, how it benchmarks, what it costs, and where it fits. For the full launch story, see our Sonnet 5 launch coverage.

What is Claude Sonnet 5?

Claude Sonnet 5 is the mid-tier model in Anthropic's Claude lineup, positioned between the fast, low-cost Haiku models and the high-capability Opus and Mythos-class models. Its API model string is claude-sonnet-5. It accepts text and image input, returns text, and is tuned specifically for agentic work: planning, tool use, coding, and knowledge tasks that run across many steps rather than a single response.

The Sonnet tier has long been the practical middle of the lineup. What sets Sonnet 5 apart is how far Anthropic pushed it toward autonomous, multi-step execution. Early testers reported that it finishes complex tasks where previous Sonnet models would stop partway, and that it checks its own output without being asked. It is a drop-in upgrade for the previous generation, Sonnet 4.6, with a few behavior changes worth knowing before you migrate.

Where Sonnet 5 fits in the Claude model family

Sonnet 5 is the high-value workhorse of the current Claude lineup, sitting below the Opus and Mythos-class models on raw capability but ahead of them on cost efficiency. The table below shows how it compares to the other current Claude models on the specifications that shape a model choice.

Current Claude model lineup by context, output, price, and speed. Source: Claude Platform Docs.

The pattern is straightforward. Fable 5 is the tier up for the hardest, longest-running agentic work. Opus 5 covers demanding coding and reasoning at a mid-high price. Haiku 4.5 is the fastest and cheapest option for high-volume, simple tasks. Sonnet 5 is the broad default: capable enough for most professional and agentic workloads, fast enough to run repeatedly, and priced to survive high volume.

What's new in Sonnet 5

Sonnet 5 is the biggest generation-over-generation leap in the Sonnet line so far, and it introduces several changes that affect how you use it. The gains span reasoning, tool use, coding, and knowledge work rather than a single category. Alongside the capability improvements, three technical changes stand out.

1. Adaptive thinking on by default

Sonnet 5 decides how much to reason on its own. Rather than a separate extended-thinking toggle, it adjusts its reasoning depth automatically based on the task in front of it. This replaces the manual extended-thinking mode that earlier models exposed, so simple prompts stay fast while harder ones get more deliberation without any configuration from you.

2. Effort control and behavior changes

You steer Sonnet 5's reasoning depth with an effort setting that defaults to high on the API and Claude Code. Two behavior changes come with the new design: manual extended thinking, which was deprecated on Sonnet 4.6, now returns an error, and setting the sampling parameters that control output randomness, such as temperature and top_p, to non-default values does the same. Code that relied on either will need a small update before it runs.

3. A new tokenizer

Sonnet 5 uses a new tokenizer, the same change Anthropic introduced with Opus 4.7. The tradeoff is that the same input can map to more tokens, roughly 1.0 to 1.35 times depending on the content type, so the real per-task cost can run higher than the sticker price suggests. If you are moving an existing workload over, recount a sample of your prompts before assuming the bill stays flat. For a deeper look at the differences, see our comparison of Sonnet 4.6 and Sonnet 5.

Claude Sonnet 5 benchmarks

Sonnet 5 posts major gains over Sonnet 4.6 across every evaluation Anthropic disclosed, lands within a few points of Opus 4.8 on most, and edges ahead of the flagship on one. The table below shows the headline results from Anthropic's launch benchmarks.

Claude Sonnet 5 benchmark results versus Sonnet 4.6 and Opus 4.8. Vendor-reported figures from Anthropic's Claude Sonnet 5 announcement and System Card.

Two results are worth calling out. On Terminal-Bench 2.1, Sonnet 5 does not just close the gap to Opus 4.8; it passes it, 80.4% to 74.6%, a jump of more than 13 points over its own predecessor. On GDPval-AA v2, a knowledge-work benchmark, Sonnet 5 edges Opus 4.8 by a slim margin, which Anthropic's System Card notes is the first time a Sonnet-class model has outscored the concurrent Opus flagship on any benchmark. Opus 4.8 still leads on the hardest coding and reasoning tasks. For a full breakdown of what each number means, read our guide to every benchmark score. All figures come from Anthropic's launch announcement and the accompanying System Card.

How Sonnet 5 compares to Opus

Sonnet 5 gives you most of Opus 4.8's capability at roughly 40 to 60 percent of the cost, which turns the old fixed-tier choice into a cost-versus-accuracy decision at the margin. According to Anthropic, Sonnet 5's performance is close to Opus 4.8, and on knowledge work it slightly leads. Opus remains the higher-capability flagship: it holds a clear lead on the hardest coding benchmark, pulls ahead on the most complex reasoning, and is the model Anthropic still recommends for cybersecurity work that Sonnet 5 was deliberately restricted from.

When to choose Sonnet 5 versus Opus 4.8 by workload, price, and speed.

The practical read is that Sonnet 5 handles the broad middle of everyday coding, agentic, and knowledge tasks, and you escalate to Opus only when a workload genuinely needs the last increment of accuracy. Our full Sonnet 5 vs Opus 4.8 comparison lays out where that crossover point sits. To see how it measures up outside the Claude family, compare Sonnet 5 and GPT-5.6.

How to access and use Claude Sonnet 5

Claude Sonnet 5 is available everywhere from launch, with no waitlist or preview gating. It is the default model for Free and Pro users, and is available to Max, Team, and Enterprise users. Developers can call it through the Claude API using the model string claude-sonnet-5, and it is available on the native Claude Platform, the Claude Platform on AWS, and Claude in Microsoft Foundry, with Google Cloud Vertex following after launch.

Real-world use cases for Claude Sonnet 5

Sonnet 5 earns its place in workflows that run frequent model calls or span many steps, where cost and follow-through both matter. Its combination of reliable tool use, long-context handling, and lower price makes it a strong default across several categories.

1. Autonomous coding agents

Autonomous coding is the clearest fit for Sonnet 5. It can take a feature request, write the code, run tests, read the failures, and iterate toward a working result while staying oriented to the original goal. Its Terminal-Bench 2.1 result, where it outscores Opus 4.8, points to the same strength: it holds up across the long, multi-step sequences that agentic coding demands.

2. Document processing at scale

Sonnet 5 handles high-volume document work at a price that makes it feasible. Legal review, contract analysis, and research summarization all benefit from its 1 million token context window, which lets it read long source material in a single pass. The lower per-token cost is what makes running that across thousands of documents practical rather than prohibitive.

3. Customer service automation

Sonnet 5 suits customer-facing automation where following instructions and staying in scope matter. It can hold a defined role, draw on product knowledge you supply, and keep responses within the boundaries you set, which is the difference between a support bot that helps and one that improvises. Its speed keeps response times low even under steady traffic.

4. Data enrichment pipelines

Sonnet 5 works well in pipelines that pull context from several sources and return structured output. It can read messy inputs, apply consistent rules, and produce clean, formatted results at the scale these jobs require. The combination of reliable output formatting and low cost per call is what makes it a fit for high-throughput enrichment.

The bottom line on Claude Sonnet 5

Claude Sonnet 5 is the strongest value in the current Claude lineup for most agentic and coding work. It brings performance close to Opus 4.8 to a Sonnet-tier price of $2 per million input tokens and $10 per million output tokens, with a 1 million token context window and adaptive thinking on by default. For the broad middle of professional workloads, it is the model to reach for first, with Opus held in reserve for the hardest tasks and Haiku for high-volume simple ones. If Sonnet 5 is not the right fit for a specific workload, our guide to Sonnet 5 alternatives covers the options across capability, cost, and architecture.

The fastest way to put Sonnet 5 to work is inside an application you can actually ship, and Claude Sonnet 5 is available on Emergent from day one. Describe what you want to build, select Sonnet 5 for your project, and go from idea to a deployed, full-stack app without managing API keys or infrastructure.

Start Building with Emergent.