Claude Sonnet 5

What it is

Claude Sonnet 5 is Anthropic’s Sonnet-tier large language model, released on 30 June 2026 and called on the Claude API as claude-sonnet-5. Anthropic describes it as “the best combination of speed and intelligence,” the balanced middle of its range, and pitches it on value: close to top-tier performance at a fraction of the top-tier price. It replaced Claude Sonnet 4.6, now listed as a legacy model, and it is the default model for users on the free and Pro plans.

Where it sits in the range

Anthropic’s line-up is not a single clean generation but a set of tiers built for different jobs. At the top are Claude Fable 5 and, since 24 July 2026, Claude Opus 5, reserved for the hardest reasoning, coding and agentic work. At the fast end is Claude Haiku 4.5. Sonnet 5 is the balance in between. Its list price is $3 per million input tokens and $15 per million output tokens, with an introductory rate of $2 and $10 running to 31 August 2026. That sits well below the Opus tier’s $5 and $25, while Anthropic’s own figures put Sonnet 5’s performance close behind it. The company’s phrase for this is “near-Opus intelligence at Sonnet pricing,” and it is the whole argument for the Sonnet tier: most of the capability for a fraction of the cost.

What it is built for

Anthropic calls Sonnet 5 “the most agentic Sonnet model yet,” able to “make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models.” It carries a one-million-token context window, so it can hold a large codebase or document set in view at once. Adaptive thinking is built in, with the reasoning effort defaulting to high on the API and in Claude Code, and there is no separate extended-thinking switch to toggle. Its training knowledge runs to January 2026. Maximum output is 128,000 tokens on the standard Messages API, rising to 300,000 on the Batch API through a beta option.

How to read the numbers

Before the scores, a caution that applies to every model page. A benchmark is a fixed set of tasks run to produce a comparable figure, and the party publishing a launch benchmark has an interest in the result, so a single number should be weighed against independent testing rather than taken at face value. On Anthropic’s own figures, Sonnet 5 scores 85.2 per cent on SWE-bench Verified, the human-checked coding benchmark, 80.4 on Terminal-Bench 2.1, and 81.2 on OSWorld-Verified for computer use. On Humanity’s Last Exam it reaches 43.2 per cent without tools and 57.4 per cent with them. These are Anthropic’s published results; the sensible course is to read them alongside the company’s model documentation and independent evaluations.

Why it stands out

The Sonnet tier is where most day-to-day work runs, because it is the point where capability and cost meet. Sonnet 5’s significance is that a mid-tier model now does agentic work, planning, tool use and autonomous runs, that a few months earlier needed a flagship. That is the direction the field has been moving: capability sliding down the price range as it matures, so the balanced tier keeps absorbing jobs that once demanded the most expensive model. What to watch is how long that holds before the next Sonnet release resets the balance again. For how it compares with the rest of the field, see our large language models hub and the wider AI section; Anthropic’s own model documentation sets out the current line in detail.