Claude Sonnet 5: Anthropic's agentic Sonnet, its price, its benchmarks and Sonnet 5.5
On 30 June 2026, Anthropic shipped Claude Sonnet 5, 'the most agentic Sonnet model yet', with a 1M-token context and a price of 2 and 10 dollars per million tokens that Anthropic later made permanent. The benchmarks from the system card, the customers, the caveats, and its successor Sonnet 5.5, released on 28 September.
In short
- Anthropic released Claude Sonnet 5 (claude-sonnet-5) on 30 June 2026: default for Free and Pro, 1M-token context, 128k output, January 2026 knowledge cutoff.
- Price: 2 dollars per million input tokens and 10 per million output. Announced as introductory until 31 August, it became the standard price on 10 August; the planned increase to 3 and 15 dollars did not happen.
- System card: SWE-bench Pro 63.2 (Sonnet 4.6: 58.1), Terminal-Bench 2.1 80.4 (67.0), and on GDPval-AA, run by Artificial Analysis, 1618, statistically tied with Opus 4.8 (1615).
- On 28 September 2026, Sonnet 5.5 replaced it at the same price, 30%+ faster and up to 30% cheaper per task according to Anthropic. Sonnet 5 is now a legacy model.
On 30 June 2026, Anthropic released Claude Sonnet 5 (API model ID claude-sonnet-5), generally available the same day: the default model for Free and Pro plans, available to Max, Team and Enterprise, and on the Claude API. Anthropic's pitch: "the most agentic Sonnet model yet", a model that plans, uses tools such as browsers and terminals and runs autonomously.
This article was first published on launch day and revised on 17 October 2026. Two things changed since: the "introductory" price of 2 and 10 dollars per million tokens became the standard price, and Claude Sonnet 5.5 replaced Sonnet 5 as the current Sonnet on 28 September. The benchmark figures below are now checked against the Sonnet 5 system card itself.
The spec sheet
| Claude Sonnet 5 | Detail |
|---|---|
| API model ID | claude-sonnet-5 (Amazon Bedrock anthropic.claude-sonnet-5, Google Cloud claude-sonnet-5) |
| Released | 30 June 2026; legacy model since Sonnet 5.5, retirement not sooner than 30 June 2027 |
| Context window | 1 million tokens |
| Max output | 128,000 tokens (300,000 on the Batch API with a beta header) |
| Knowledge cutoff | January 2026 |
| Input and output | Text and images in, text out |
| Thinking | Adaptive thinking; effort defaults to high on the API and in Claude Code |
| Source: Claude Sonnet 5 documentation, read on 17 October 2026. | |
The price: the introductory rate stayed
At launch, Anthropic announced 2 dollars per million input tokens and 10 dollars per million output tokens as an introductory price until 31 August 2026, with 3 and 15 dollars to follow on 1 September. On 10 August, it changed course: the 2 and 10 dollar price is permanent, and the increase "will not occur" (announcement, edited on 10 August; price list).
| Model (Anthropic price list, 17 October 2026) | Input / output per million tokens | Cache read |
|---|---|---|
| Claude Haiku 5.5 | 0.10 / 0.50 dollar (prompts up to 100,000 tokens) | 0.01 dollar |
| Claude Sonnet 5 | 2 / 10 dollars | 0.20 dollar |
| Claude Sonnet 5.5 | 2 / 10 dollars | 0.10 dollar |
| Claude Opus 4.8 | 5 / 25 dollars | 0.50 dollar |
| Claude Opus 5.5 | 4 / 20 dollars | 0.20 dollar |
| Claude Fable 5.1 | 10 / 50 dollars | 0.25 dollar |
One caveat for budgets: models from Claude Opus 4.7 onwards, Sonnet 5 included, use a tokenizer that produces about 30% more tokens for the same text than Sonnet 4.6 and earlier, according to the same price list. Measure your own workload rather than comparing list prices alone.
The benchmarks: a generational jump over Sonnet 4.6
The figures below come from the evaluation summary of the system card. They are reported by Anthropic, except GDPval-AA, which the card says was run independently by Artificial Analysis.
| Evaluation (system card) | Sonnet 5 | Sonnet 4.6 | GPT-5.5 | Gemini 3.5 Flash |
|---|---|---|---|---|
| SWE-bench Pro (agentic coding) | 63.2 | 58.1 | 58.6 | 55.1 |
| Terminal-Bench 2.1 | 80.4 | 67.0 | 83.4 (Codex CLI) | 76.2 |
| Humanity's Last Exam, no tools | 43.2 | 34.6 | 41.4 | 40.2 |
| Humanity's Last Exam, with tools | 57.4 | 46.8 | 52.2 | - |
| OSWorld-Verified (computer use) | 81.2 | 78.5 | 78.7 | 78.4 |
| FrontierCode v1 | 38.8 | 15.1 | 25.5 | - |
| GDPval-AA v2 (knowledge work, Elo) | 1618 | 1395 | 1509 | 1357 |
The row that sums up the release is the last one. On GDPval-AA, the card says Sonnet 5 "ranks second (ELO 1618), statistically tied with Opus 4.8 (ELO 1615) and trailing only Fable 5 (ELO 1783)", at less than half Opus 4.8's price. The Terminal-Bench jump, from 67 to 80.4 in one Sonnet generation, is the other striking figure: real progress on running commands in a real shell, which is what agents do. On the same test, GPT-5.5 in Codex CLI stayed ahead.
Who used it on day one
The announcement quotes named customers. Among them: Lovable ("gets more done with less. Same output quality, fewer steps to get there"), ClickHouse ("reasons in tighter steps and gets our users to answers noticeably faster"), Eve ("sits on the Pareto frontier for Eve's plaintiff-law tasks") and Pace ("consistently takes the right action and does it quickly").
What happened since: Sonnet 5.5
On 28 September 2026, Anthropic released Claude Sonnet 5.5 (claude-sonnet-5-5), "a clear upgrade over Claude Sonnet 5", which "runs 30%+ faster, and costs up to 30% less for most work". The list price is the same; the saving comes from fewer tokens per task and cache reads halved to 0.10 dollar. Its comparison table uses newer versions of several benchmarks, which is why Sonnet 5's scores there differ from the ones above: for example 64.5% against 54.9% on Humanity's Last Exam with tools, and 1844 against 1449 on GDPval-AA v2.1. Anthropic's documentation now describes Sonnet 5 as a legacy model and recommends migrating.
The caveats
- Most benchmarks are Anthropic's own. Only GDPval-AA was run by a third party, according to the card.
- The tokenizer changes the bill. About 30% more tokens for the same text than Sonnet 4.6: a migration from Sonnet 4.6 is not a straight price comparison.
- "Most agentic Sonnet yet" is Anthropic's framing. The data supports a strong model; the superlative is not an independent verdict.
Our read
What follows is our analysis.
In July, Sonnet 5 was the sensible default for most coding and agentic work: close to Opus 4.8 on knowledge work, well ahead of Sonnet 4.6 on agentic coding, at 2 and 10 dollars, a price Anthropic then kept. In October, the same reasoning points to Sonnet 5.5, which costs the same, runs faster and uses fewer tokens. Keep Sonnet 5 only where you have validated it and cannot re-test yet.
The deeper point is that the model is a replaceable component, not the product. Put it behind your own interface, so that moving from Sonnet 4.6 to 5, then to 5.5, is a configuration change and a test run, not a rebuild. That is how we build our own products. For the rest of that summer's model wave, read the Fable 5 suspension, Grok 4.5 and GLM-5.2.
Frequently asked questions
What is Claude Sonnet 5?
The Sonnet model Anthropic released on 30 June 2026 under the API ID claude-sonnet-5, presented as the most agentic Sonnet model yet. It has a 1M-token context window, 128k tokens of maximum output, a January 2026 knowledge cutoff, image input and adaptive thinking. It was the default model for Free and Pro plans at launch.
How much does Claude Sonnet 5 cost?
2 dollars per million input tokens and 10 dollars per million output tokens, with cache reads at 0.20 dollar. This was announced as introductory pricing until 31 August 2026, but on 10 August Anthropic made it the standard price and cancelled the planned move to 3 and 15 dollars. The Batch API is 50% off.
Is Claude Sonnet 5 as good as Opus 4.8?
On knowledge work, close: on GDPval-AA, run independently by Artificial Analysis, Sonnet 5 scores 1618 against 1615 for Opus 4.8, a statistical tie according to the system card, and Fable 5 leads with 1783. Opus 4.8 costs 5 and 25 dollars per million tokens against 2 and 10 for Sonnet 5.
What is the context window and knowledge cutoff of Claude Sonnet 5?
A 1-million-token context window, 128,000 tokens of maximum output (300,000 on the Batch API with a beta header), and a reliable knowledge cutoff of January 2026, according to Anthropic's documentation.
Should I use Claude Sonnet 5 or Sonnet 5.5?
Sonnet 5.5, released on 28 September 2026 at the same list price: Anthropic says it runs more than 30% faster and costs up to 30% less for most work, and its cache reads cost half. Sonnet 5 remains available as a legacy model, with retirement not sooner than 30 June 2027.
Does Sonnet 5 cost more tokens than Sonnet 4.6?
Yes, for the same text: Anthropic's price list says models from Opus 4.7 onwards use a tokenizer that produces about 30% more tokens than Sonnet 4.6 and earlier. Compare costs on your own workload, not on list prices alone.
Read next
Is your project decided?
Present it in writing. We read everything, then we answer.
Pitch a project